2023-03-10 19:47:46 +00:00
# llama.cpp
2023-04-05 15:56:20 +00:00
2023-03-26 07:20:49 +00:00
2023-03-12 20:09:26 +00:00
[](https://opensource.org/licenses/MIT)
2023-10-04 13:50:44 +00:00
[Roadmap ](https://github.com/users/ggerganov/projects/7 ) / [Project status ](https://github.com/ggerganov/llama.cpp/discussions/3471 ) / [Manifesto ](https://github.com/ggerganov/llama.cpp/discussions/205 ) / [ggml ](https://github.com/ggerganov/ggml )
2023-06-25 13:08:12 +00:00
2024-02-05 14:55:10 +00:00
Inference of Meta's [LLaMA ](https://arxiv.org/abs/2302.13971 ) model (and others) in pure C/C++
2023-03-10 19:47:46 +00:00
2024-03-03 10:44:03 +00:00
### Recent API changes
2024-04-21 15:36:45 +00:00
- [2024 Apr 21] `llama_token_to_piece` can now optionally render special tokens https://github.com/ggerganov/llama.cpp/pull/6807
2024-04-08 12:43:30 +00:00
- [2024 Apr 4] State and session file functions reorganized under `llama_state_*` https://github.com/ggerganov/llama.cpp/pull/6341
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
- [2024 Mar 26] Logits and embeddings API updated for compactness https://github.com/ggerganov/llama.cpp/pull/6122
2024-03-13 18:33:56 +00:00
- [2024 Mar 13] Add `llama_synchronize()` + `llama_context_params.n_ubatch` https://github.com/ggerganov/llama.cpp/pull/6017
2024-03-11 15:49:47 +00:00
- [2024 Mar 8] `llama_kv_cache_seq_rm()` returns a `bool` instead of `void` , and new `llama_n_seq_max()` returns the upper limit of acceptable `seq_id` in batches (relevant when dealing with multiple sequences) https://github.com/ggerganov/llama.cpp/pull/5328
2024-03-04 20:31:20 +00:00
- [2024 Mar 4] Embeddings API updated https://github.com/ggerganov/llama.cpp/pull/5796
2024-03-03 10:44:03 +00:00
- [2024 Mar 3] `struct llama_context_params` https://github.com/ggerganov/llama.cpp/pull/5849
2023-08-21 20:07:43 +00:00
### Hot topics
2023-08-27 11:44:35 +00:00
2024-05-07 18:43:13 +00:00
- **Initial Flash-Attention support: https://github.com/ggerganov/llama.cpp/pull/5021**
- BPE pre-tokenization support has been added: https://github.com/ggerganov/llama.cpp/pull/6920
2024-04-29 14:06:19 +00:00
- MoE memory layout has been updated - reconvert models for `mmap` support and regenerate `imatrix` https://github.com/ggerganov/llama.cpp/pull/6387
2024-03-31 08:56:30 +00:00
- Model sharding instructions using `gguf-split` https://github.com/ggerganov/llama.cpp/discussions/6404
2024-03-22 09:35:53 +00:00
- Fix major bug in Metal batched inference https://github.com/ggerganov/llama.cpp/pull/6225
2024-04-04 17:16:37 +00:00
- Multi-GPU pipeline parallelism support https://github.com/ggerganov/llama.cpp/pull/6017
2024-03-10 18:58:26 +00:00
- Looking for contributions to add Deepseek support: https://github.com/ggerganov/llama.cpp/issues/5981
- Quantization blind testing: https://github.com/ggerganov/llama.cpp/discussions/5962
2024-03-09 16:14:13 +00:00
- Initial Mamba support has been added: https://github.com/ggerganov/llama.cpp/pull/5328
2023-08-18 14:48:31 +00:00
----
2023-03-14 07:43:52 +00:00
2023-05-08 16:33:30 +00:00
< details >
< summary > Table of Contents< / summary >
< ol >
< li >
< a href = "#description" > Description< / a >
< / li >
< li >
< a href = "#usage" > Usage< / a >
< ul >
< li > < a href = "#get-the-code" > Get the Code< / a > < / li >
< li > < a href = "#build" > Build< / a > < / li >
< li > < a href = "#blas-build" > BLAS Build< / a > < / li >
2024-02-07 06:21:30 +00:00
< li > < a href = "#prepare-and-quantize" > Prepare and Quantize< / a > < / li >
< li > < a href = "#run-the-quantized-model" > Run the quantized model< / a > < / li >
2023-05-08 16:33:30 +00:00
< li > < a href = "#memorydisk-requirements" > Memory/Disk Requirements< / a > < / li >
< li > < a href = "#quantization" > Quantization< / a > < / li >
< li > < a href = "#interactive-mode" > Interactive mode< / a > < / li >
2023-08-23 01:01:57 +00:00
< li > < a href = "#constrained-output-with-grammars" > Constrained output with grammars< / a > < / li >
2024-02-07 06:21:30 +00:00
< li > < a href = "#instruct-mode" > Instruct mode< / a > < / li >
< li > < a href = "#obtaining-and-using-the-facebook-llama-2-model" > Obtaining and using the Facebook LLaMA 2 model< / a > < / li >
2023-05-08 16:33:30 +00:00
< li > < a href = "#seminal-papers-and-background-on-the-models" > Seminal papers and background on the models< / a > < / li >
< li > < a href = "#perplexity-measuring-model-quality" > Perplexity (measuring model quality)< / a > < / li >
< li > < a href = "#android" > Android< / a > < / li >
< li > < a href = "#docker" > Docker< / a > < / li >
< / ul >
< / li >
< li > < a href = "#contributing" > Contributing< / a > < / li >
< li > < a href = "#coding-guidelines" > Coding guidelines< / a > < / li >
< li > < a href = "#docs" > Docs< / a > < / li >
< / ol >
< / details >
2023-03-10 19:47:46 +00:00
## Description
2024-02-05 14:55:10 +00:00
The main goal of `llama.cpp` is to enable LLM inference with minimal setup and state-of-the-art performance on a wide
variety of hardware - locally and in the cloud.
2023-03-10 19:47:46 +00:00
2024-02-05 14:55:10 +00:00
- Plain C/C++ implementation without any dependencies
- Apple silicon is a first-class citizen - optimized via ARM NEON, Accelerate and Metal frameworks
2023-05-05 14:43:36 +00:00
- AVX, AVX2 and AVX512 support for x86 architectures
2024-02-19 07:39:31 +00:00
- 1.5-bit, 2-bit, 3-bit, 4-bit, 5-bit, 6-bit, and 8-bit integer quantization for faster inference and reduced memory use
2024-02-05 14:55:10 +00:00
- Custom CUDA kernels for running LLMs on NVIDIA GPUs (support for AMD GPUs via HIP)
- Vulkan, SYCL, and (partial) OpenCL backend support
- CPU+GPU hybrid inference to partially accelerate models larger than the total VRAM capacity
2023-03-10 19:47:46 +00:00
2024-02-05 14:55:10 +00:00
Since its [inception ](https://github.com/ggerganov/llama.cpp/issues/33#issuecomment-1465108022 ), the project has
improved significantly thanks to many contributions. It is the main playground for developing new features for the
[ggml ](https://github.com/ggerganov/ggml ) library.
2023-03-11 10:31:21 +00:00
2023-04-05 15:56:20 +00:00
**Supported platforms:**
2023-03-12 06:41:54 +00:00
- [X] Mac OS
- [X] Linux
2023-03-13 17:21:51 +00:00
- [X] Windows (via CMake)
2023-03-17 09:47:06 +00:00
- [X] Docker
2024-02-05 14:55:10 +00:00
- [X] FreeBSD
2023-03-12 06:41:54 +00:00
2023-04-05 15:56:20 +00:00
**Supported models:**
2023-03-29 16:37:20 +00:00
2024-02-07 06:21:30 +00:00
Typically finetunes of the base models below are supported as well.
2023-03-30 19:31:54 +00:00
- [X] LLaMA 🦙
2023-07-28 01:14:11 +00:00
- [x] LLaMA 2 🦙🦙
2024-04-25 13:52:28 +00:00
- [x] LLaMA 3 🦙🦙🦙
2024-02-07 06:21:30 +00:00
- [X] [Mistral 7B ](https://huggingface.co/mistralai/Mistral-7B-v0.1 )
2024-02-05 14:55:10 +00:00
- [x] [Mixtral MoE ](https://huggingface.co/models?search=mistral-ai/Mixtral )
2024-04-13 09:33:52 +00:00
- [x] [DBRX ](https://huggingface.co/databricks/dbrx-instruct )
2024-04-21 12:35:40 +00:00
- [X] [Falcon ](https://huggingface.co/models?search=tiiuae/falcon )
2023-08-02 06:18:31 +00:00
- [X] [Chinese LLaMA / Alpaca ](https://github.com/ymcui/Chinese-LLaMA-Alpaca ) and [Chinese LLaMA-2 / Alpaca-2 ](https://github.com/ymcui/Chinese-LLaMA-Alpaca-2 )
2023-03-30 19:31:54 +00:00
- [X] [Vigogne (French) ](https://github.com/bofenghuang/vigogne )
2023-04-10 20:41:53 +00:00
- [X] [Koala ](https://bair.berkeley.edu/blog/2023/04/03/koala/ )
2023-10-17 18:13:21 +00:00
- [X] [Baichuan 1 & 2 ](https://huggingface.co/models?search=baichuan-inc/Baichuan ) + [derivations ](https://huggingface.co/hiyouga/baichuan-7b-sft )
- [X] [Aquila 1 & 2 ](https://huggingface.co/models?search=BAAI/Aquila )
2023-09-29 12:50:35 +00:00
- [X] [Starcoder models ](https://github.com/ggerganov/llama.cpp/pull/3187 )
2023-10-06 19:13:36 +00:00
- [X] [Refact ](https://huggingface.co/smallcloudai/Refact-1_6B-fim )
2023-10-17 18:13:21 +00:00
- [X] [Persimmon 8B ](https://github.com/ggerganov/llama.cpp/pull/3410 )
2023-10-10 23:02:49 +00:00
- [X] [MPT ](https://github.com/ggerganov/llama.cpp/pull/3417 )
2023-10-17 18:13:21 +00:00
- [X] [Bloom ](https://github.com/ggerganov/llama.cpp/pull/3553 )
2023-12-14 07:38:49 +00:00
- [x] [Yi models ](https://huggingface.co/models?search=01-ai/Yi )
2024-02-06 14:06:48 +00:00
- [X] [StableLM models ](https://huggingface.co/stabilityai )
2023-12-14 07:38:49 +00:00
- [x] [Deepseek models ](https://huggingface.co/models?search=deepseek-ai/deepseek )
- [x] [Qwen models ](https://huggingface.co/models?search=Qwen/Qwen )
2023-12-24 13:35:49 +00:00
- [x] [PLaMo-13B ](https://github.com/ggerganov/llama.cpp/pull/3557 )
2024-02-06 14:06:48 +00:00
- [x] [Phi models ](https://huggingface.co/models?search=microsoft/phi )
2023-12-28 14:03:57 +00:00
- [x] [GPT-2 ](https://huggingface.co/gpt2 )
2024-02-06 14:06:48 +00:00
- [x] [Orion 14B ](https://github.com/ggerganov/llama.cpp/pull/5118 )
- [x] [InternLM2 ](https://huggingface.co/models?search=internlm2 )
2024-02-05 07:41:38 +00:00
- [x] [CodeShell ](https://github.com/WisdomShell/codeshell )
2024-02-21 13:08:22 +00:00
- [x] [Gemma ](https://ai.google.dev/gemma )
llama : support Mamba Selective State Space Models (#5328)
* mamba : begin working on support for Mamba SSM
* mamba : begin figuring out how to (ab)use the kv cache for Mamba
* mamba : recurrent inference almost works, but incoherent
* mamba : recurrent inference WORKS!!!
* convert : optionally use d_conv and d_state from config.json for Mamba
* mamba : refactor recurrent conv, resulting in 20% perf increase
It's still slower than I'd like, but I did not really optimize `ggml_exp` yet.
I also refactored `ggml_exp` to work with tensors with more than 2 dimensions.
* ggml : parallelize ggml_exp
This results in 8% faster token generation for Mamba-130M.
* mamba : simplify the conv step with a self-overlapping view
Turns out the conv_state can be made smaller by one column.
Note that this breaks existing GGUFs of Mamba,
because the key_value_length field is tied to the conv_state size.
Convolution with a self-overlapping view is cool!
And it's much simpler than what I initially thought would be necessary
to make the convolution step work with more than 1 token at a time.
Next step is to make the SSM step work on batches of tokens too,
and thus I need to figure out a way to make a parallel selective scan
which will keep the ssm_state small and won't make it bigger
by a factor of (n_layer * batch_size).
* llama : fix Mamba KV self size wrongly displaying as f16 instead of f32
Relatedly, I also tried to see if other types than f32 worked for the states,
but they don't, because of the operators used.
It's probably better anyway to keep lots of precision there,
since the states are small anyway.
* mamba : fix self-overlapping view depth stride
* mamba : handle batches of more than 1 token
This means running Mamba no longer crashes when using the default settings!
And probably also slightly faster prompt processing.
Both batched and non-batched processing yield the same output.
Previously, the state was not cleared when starting a sequence.
Next step is to make the KV cache API work as expected for Mamba models.
* ggml: add ggml_ssm_scan to help with parallel selective scan
If the selective scan was implemented without a custom operator,
there would be waaay too many nodes in the graph. For example,
for Mamba-130M, with a batch size of 512 (the default),
a naive selective scan could add at least 24*512=12288 nodes,
which is more than LLAMA_MAX_NODES (8192),
and that's only for the smallest Mamba model.
So it's much cleaner with a custom operator.
Not sure about the name, though.
* ggml : in ggml_ssm_scan, merge multiple rows in the same vec operation
This will help with performance on CPU if ggml_vec_mul_f32
and ggml_vec_add_f32 are ever optimized with SIMD.
* mamba : very basic quantization support
Mostly works, but there is currently no difference
between the variants of a k-quant (e.g. Q4_K_S and Q4_K_M are the same).
Most of the SSM-specific weights can be kept in f32 without affecting
the size that much, since they are relatively small.
(the linear projection weights are responsible for most of Mamba's size)
Too much quantization seems to make the state degrade quite fast, and
the model begins to output gibberish.
It seems to affect bigger models to a lesser extent than small models,
but I'm not sure by how much.
Experimentation will be needed to figure out which weights are more important
for the _M (and _L?) variants of k-quants for Mamba.
* convert : fix wrong name for layer norm weight of offical Mamba models
I was using Q-bert/Mamba-* models before, which have a slighlty different
naming scheme for the weights.
(they start with "model.layers" instead of "backbone.layers")
* mamba : fuse more steps of the SSM scan in the ggml_ssm_scan operator
This increases performance on CPU by around 30% for prompt processing,
and by around 20% for text generation.
However, it also makes the ggml_exp and ggml_soft_plus operators unused.
Whether or not they should be kept will be decided later.
* convert : for Mamba, also consider the "MambaLMHeadModel" arch name
It's the name of the class of the official implementation,
though they don't use it (yet) in the "architectures" field of config.json
* mamba : fix vocab size problems with official models
The perplexity was waaaay to high for models with a non-round vocab size.
Not sure why, but it needed to be fixed in the metadata.
Note that this breaks existing GGUF-converted Mamba models,
but **only if** the vocab size was not already rounded.
* ggml : remove ggml_exp and ggml_soft_plus
They did not exist anyway outside of this branch,
and since ggml_ssm_scan fused operations together, they are unused.
It's always possible to bring them back if needed.
* mamba : remove some useless comments
No code change.
* convert : fix flake8 linter errors
* mamba : apply suggestions from code review
* mamba : remove unecessary branch for row-wise ssm_state and C multiplication
It was previously done to avoid permuting when only one token is processed
at a time (like when generating text), but permuting is cheap,
and dynamically changing the compute graph is not future-proof.
* ggml : in ggml_ssm_scan, use more appropriate asserts
* ggml : rename the destination pointer in ggml_compute_forward_ssm_scan_f32
* mamba : multiple sequences, but one at a time
This is a step towards making this Mamba implementation usable
with the server example (the way the system prompt is kept when clearing
the client slots will need to be changed before this can work, though).
The KV cache size for this kind of model is tied to the maximum number
of sequences kept at any single time.
For now, this number is obtained from n_parallel (plus one,
to have an extra sequence to dedicate to the system prompt),
but there might be a better way to do this which won't also
make the main example use 2 cells even if only 1 is really used.
(for this specific case, --parallel 0 helps)
Simultaneous sequence processing will probably require changes to
ggml_ssm_scan, and possibly a new operator for the conv step.
* mamba : support llama_kv_cache_seq_cp
This (mis)uses the logic around K shifts, because tokens in a state
can't be shifted anyway, and because inp_K_shift has the right shape and type.
Using ggml_get_rows is a nice way to do copies, but copy chains can't work.
Fortunately, copy chains don't really seem to be used in the examples.
Each KV cell is dedicated to the sequence ID corresponding to its own index.
* mamba : use a state mask
It's cleaner than the previous heuristic of
checking for the pos of the first token in the batch.
inp_KQ_mask could not be re-used for this, because it has the wrong shape
and because it seems more suited to the next step of
simultaneous sequence processing (helping with the problem of
remembering which token belongs to which sequence(s)/state(s)).
* llama : replace the usage of n_ctx with kv_self.size in many places
* mamba : use n_tokens directly instead of n_tok
* mamba : in comments, properly refer to KV cells instead of slots
* mamba : reduce memory usage of ggml_ssm_scan
From 290.37 MiB to 140.68 MiB of CPU compute buffer size
with Mamba 3B with a batch size of 512.
The result tensor of ggml_ssm_scan was previously a big part
of the CPU compute buffer size. To make it smaller,
it does not contain the intermediate ssm states anymore.
Both y and the last ssm state are combined in the result tensor,
because it seems only a single tensor can be returned by an operator
with the way the graph is built.
* mamba : simultaneous sequence processing
A batch can now contain tokens from multiple sequences.
This is necessary for at least the parallel example, the server example,
and the HellaSwag test in the perplexity example.
However, for this to be useful, uses of llama_kv_cache_seq_rm/cp
will need to be changed to work on whole sequences.
* ggml : add ggml_ssm_conv as a new operator for the conv step of Mamba
This operator makes it possible to use and update the correct states
for each token of the batch in the same way as ggml_ssm_scan.
Other solutions which use existing operators would need loops which would
add too many nodes to the graph (at least the ones I thought of).
Using this operator further reduces the size of the CPU compute buffer
from 140.68 MiB to 103.20 MiB with Mamba 3B with a batch size of 512.
And (at least on CPU), it's a bit faster than before.
Note that "ggml_ssm_conv" is probably not the most appropriate name,
and it could be changed if a better one is found.
* llama : add inp_s_seq as a new input tensor
The most convenient implementation to select the correct state (for Mamba)
for each token is to directly get the correct index from a tensor.
This is why inp_s_seq is storing int32_t and not floats.
The other, less convenient way to select the correct state would be
to have inp_KQ_mask contain 1.0f for each state used by a token
and 0.0f otherwise. This complicates quickly fetching the first used
state of a token, and is also less efficient because a whole row
of the mask would always need to be read for each token.
Using indexes makes it easy to stop searching when there are
no more sequences for a token, and the first sequence assigned
is always very quickly available (it's the first element of each row).
* mamba : support llama_kv_cache_seq_cp copy chains
* mamba : support shifting and dividing the kv cache pos
* mamba : make the server and parallel examples work with whole sequences
A seq_id is dedicated to the system prompt in both cases.
* llama : make llama_kv_cache_seq_rm return whether it succeeded or not
* mamba : dedicate an input tensor for state copy indices
This is cleaner and makes it easier to adapt when/if token positions
(and by extension, inp_K_shift) are no longer integers.
* mamba : adapt perplexity, batched, and batched-bench examples
* perplexity : limit the max number of sequences
This adapts to what the loaded model can provide.
* llama : add llama_n_max_seq to get the upper limit for seq_ids
Used by the perplexity example.
* batched : pass n_parallel to the model's context params
This should have been there already, but it wasn't.
* batched-bench : reserve sequences to support Mamba
* batched-bench : fix tokens being put in wrong sequences
Generation quality isn't what's measured in there anyway,
but at least using the correct sequences avoids using non-consecutive
token positions.
* mamba : stop abusing attention metadata
This breaks existing converted-to-GGUF Mamba models,
but will allow supporting mixed architectures like MambaFormer
without needing to break Mamba models.
This will also allow changing the size of Mamba's states
without having to reconvert models in the future.
(e.g. using something else than d_conv - 1 columns for the conv_states
will not require breaking existing converted Mamba models again)
* gguf-py : add new KV metadata key-value pairs for Mamba
* llama : add new metadata key-value pairs for Mamba
* llama : guard against divisions by zero when n_head is 0
* mamba : rename "unlimited" KV cache property to "recurrent"
* mamba : more correctly update the "used" field of the KV cache
* ggml : in ggml_ssm_scan, use a threshold for soft_plus
This is how the official Mamba implementation does it,
and it's also what torch.nn.Softplus does.
* convert : for Mamba, fallback to internal NeoX tokenizer
The resulting models are exactly the same
as if the tokenizer.json and tokenizer_config.json of GPT-NeoX were there.
* mamba : support state saving and restoring
* ggml : implicitly pass src tensors through dst for Mamba-related ops
* mamba : clarify some comments
* server : fix cache_tokens not getting correctly resized
Otherwise, when the "we have to evaluate at least 1 token" special case
was triggered, an extra token was kept in cache_tokens even if it was
removed from the KV cache.
For Mamba, this caused useless prompt reprocessing when the previous
request triggered the above case.
* convert-hf : support new metadata keys for Mamba
For the models available at
https://huggingface.co/collections/state-spaces/transformers-compatible-mamba-65e7b40ab87e5297e45ae406
* mamba : rename metadata to be more similar to transformers library
This breaks existing converted-to-GGUF models,
but the metadata names are more "standard".
* mamba : support mamba-*-hf models
These models share their token_embd.weight with their output.weight
* mamba : add missing spaces
This is purely a formatting change.
* convert-hf : omit output.weight when identical with token_embd.weight
Only for Mamba for now, but it might be relevant for other models eventually.
Most Mamba models actually share these two tensors, albeit implicitly.
* readme : add Mamba to supported models, and add recent API changes
* mamba : move state_seq and state_mask views outside layer loop
A few tensors were also missing `struct` in front of `ggml_tensor`.
2024-03-08 22:31:00 +00:00
- [x] [Mamba ](https://github.com/state-spaces/mamba )
2024-04-25 13:52:28 +00:00
- [x] [Grok-1 ](https://huggingface.co/keyfan/grok-1-hf )
2024-03-29 13:37:03 +00:00
- [x] [Xverse ](https://huggingface.co/models?search=xverse )
2024-04-25 13:52:28 +00:00
- [x] [Command-R models ](https://huggingface.co/models?search=CohereForAI/c4ai-command-r )
2024-04-03 18:05:10 +00:00
- [x] [SEA-LION ](https://huggingface.co/models?search=sea-lion )
2024-04-07 17:33:59 +00:00
- [x] [GritLM-7B ](https://huggingface.co/GritLM/GritLM-7B ) + [GritLM-8x7B ](https://huggingface.co/GritLM/GritLM-8x7B )
2024-04-19 09:35:54 +00:00
- [x] [OLMo ](https://allenai.org/olmo )
2023-12-14 07:38:49 +00:00
2024-04-10 06:58:48 +00:00
(instructions for supporting more models: [HOWTO-add-model.md ](./docs/HOWTO-add-model.md ))
2023-12-14 07:38:49 +00:00
**Multimodal models:**
2024-02-28 08:39:39 +00:00
- [x] [LLaVA 1.5 models ](https://huggingface.co/collections/liuhaotian/llava-15-653aac15d994e992e2677a7e ), [LLaVA 1.6 models ](https://huggingface.co/collections/liuhaotian/llava-16-65b9e40155f60fd046a5ccf2 )
2024-02-05 14:55:10 +00:00
- [x] [BakLLaVA ](https://huggingface.co/models?search=SkunkworksAI/Bakllava )
2023-12-14 07:38:49 +00:00
- [x] [Obsidian ](https://huggingface.co/NousResearch/Obsidian-3B-V0.5 )
- [x] [ShareGPT4V ](https://huggingface.co/models?search=Lin-Chen/ShareGPT4V )
2024-01-25 20:14:32 +00:00
- [x] [MobileVLM 1.7B/3B models ](https://huggingface.co/models?search=mobileVLM )
2024-02-06 14:06:48 +00:00
- [x] [Yi-VL ](https://huggingface.co/models?search=Yi-VL )
2024-04-25 13:52:28 +00:00
- [x] [Mini CPM ](https://huggingface.co/models?search=MiniCPM )
2024-04-25 19:38:31 +00:00
- [x] [Moondream ](https://huggingface.co/vikhyatk/moondream2 )
2023-10-17 18:13:21 +00:00
2024-02-25 20:46:29 +00:00
**HTTP server**
[llama.cpp web server ](./examples/server ) is a lightweight [OpenAI API ](https://github.com/openai/openai-openapi ) compatible HTTP server that can be used to serve local models and easily connect them to existing clients.
2023-04-05 15:56:20 +00:00
**Bindings:**
- Python: [abetlen/llama-cpp-python ](https://github.com/abetlen/llama-cpp-python )
- Go: [go-skynet/go-llama.cpp ](https://github.com/go-skynet/go-llama.cpp )
2023-10-22 18:16:43 +00:00
- Node.js: [withcatai/node-llama-cpp ](https://github.com/withcatai/node-llama-cpp )
2024-01-07 20:24:11 +00:00
- JS/TS (llama.cpp server client): [lgrammel/modelfusion ](https://modelfusion.dev/integration/model-provider/llamacpp )
2024-02-09 10:17:00 +00:00
- JavaScript/Wasm (works in browser): [tangledgroup/llama-cpp-wasm ](https://github.com/tangledgroup/llama-cpp-wasm )
2024-03-16 15:42:08 +00:00
- Typescript/Wasm (nicer API, available on npm): [ngxson/wllama ](https://github.com/ngxson/wllama )
2023-04-17 19:34:35 +00:00
- Ruby: [yoshoku/llama_cpp.rb ](https://github.com/yoshoku/llama_cpp.rb )
2024-04-03 17:53:37 +00:00
- Rust (more features): [edgenai/llama_cpp-rs ](https://github.com/edgenai/llama_cpp-rs )
2024-01-28 08:30:44 +00:00
- Rust (nicer API): [mdrokz/rust-llama.cpp ](https://github.com/mdrokz/rust-llama.cpp )
- Rust (more direct bindings): [utilityai/llama-cpp-rs ](https://github.com/utilityai/llama-cpp-rs )
2023-05-12 05:39:40 +00:00
- C#/.NET: [SciSharp/LLamaSharp ](https://github.com/SciSharp/LLamaSharp )
2023-06-26 19:47:59 +00:00
- Scala 3: [donderom/llm4s ](https://github.com/donderom/llm4s )
2023-08-18 19:39:22 +00:00
- Clojure: [phronmophobic/llama.clj ](https://github.com/phronmophobic/llama.clj )
2023-08-29 09:30:10 +00:00
- React Native: [mybigday/llama.rn ](https://github.com/mybigday/llama.rn )
2023-09-01 13:36:14 +00:00
- Java: [kherud/java-llama.cpp ](https://github.com/kherud/java-llama.cpp )
2023-12-22 06:49:54 +00:00
- Zig: [deins/llama.cpp.zig ](https://github.com/Deins/llama.cpp.zig )
2024-01-20 08:05:43 +00:00
- Flutter/Dart: [netdur/llama_cpp_dart ](https://github.com/netdur/llama_cpp_dart )
2024-03-27 07:08:59 +00:00
- PHP (API bindings and features built on top of llama.cpp): [distantmagic/resonance ](https://github.com/distantmagic/resonance ) [(more info) ](https://github.com/ggerganov/llama.cpp/pull/6326 )
2023-04-05 15:56:20 +00:00
**UI:**
2024-02-05 14:55:10 +00:00
Unless otherwise noted these projects are open-source with permissive licensing:
- [iohub/collama ](https://github.com/iohub/coLLaMA )
- [janhq/jan ](https://github.com/janhq/jan ) (AGPL)
2023-04-05 15:56:20 +00:00
- [nat/openplayground ](https://github.com/nat/openplayground )
2024-02-07 06:16:48 +00:00
- [Faraday ](https://faraday.dev/ ) (proprietary)
2024-02-05 14:55:10 +00:00
- [LMStudio ](https://lmstudio.ai/ ) (proprietary)
2024-02-21 14:39:10 +00:00
- [LocalAI ](https://github.com/mudler/LocalAI ) (MIT)
2024-02-05 14:55:10 +00:00
- [LostRuins/koboldcpp ](https://github.com/LostRuins/koboldcpp ) (AGPL)
- [Mozilla-Ocho/llamafile ](https://github.com/Mozilla-Ocho/llamafile )
- [nomic-ai/gpt4all ](https://github.com/nomic-ai/gpt4all )
- [ollama/ollama ](https://github.com/ollama/ollama )
- [oobabooga/text-generation-webui ](https://github.com/oobabooga/text-generation-webui ) (AGPL)
2023-11-29 07:16:34 +00:00
- [psugihara/FreeChat ](https://github.com/psugihara/FreeChat )
2024-02-07 18:44:52 +00:00
- [cztomsik/ava ](https://github.com/cztomsik/ava ) (MIT)
2023-12-25 16:09:53 +00:00
- [ptsochantaris/emeltal ](https://github.com/ptsochantaris/emeltal )
2024-02-05 14:55:10 +00:00
- [pythops/tenere ](https://github.com/pythops/tenere ) (AGPL)
2024-03-22 11:29:49 +00:00
- [RecurseChat ](https://recurse.chat/ ) (proprietary)
2024-02-05 14:55:10 +00:00
- [semperai/amica ](https://github.com/semperai/amica )
- [withcatai/catai ](https://github.com/withcatai/catai )
2024-02-20 10:00:23 +00:00
- [Mobile-Artificial-Intelligence/maid ](https://github.com/Mobile-Artificial-Intelligence/maid ) (MIT)
2024-02-25 15:57:34 +00:00
- [Msty ](https://msty.app ) (proprietary)
2024-02-26 14:15:28 +00:00
- [LLMFarm ](https://github.com/guinmoon/LLMFarm?tab=readme-ov-file ) (MIT)
2024-03-29 07:33:46 +00:00
- [KanTV ](https://github.com/zhouwg/kantv?tab=readme-ov-file )(Apachev2.0 or later)
2024-04-04 17:22:50 +00:00
- [Dot ](https://github.com/alexpinel/Dot ) (GPL)
2024-04-05 18:39:43 +00:00
- [MindMac ](https://mindmac.app ) (proprietary)
2024-04-08 07:48:29 +00:00
- [KodiBot ](https://github.com/firatkiral/kodibot ) (GPL)
2024-04-10 06:34:00 +00:00
- [eva ](https://github.com/ylsdamxssjxxdd/eva ) (MIT)
2024-04-17 12:47:50 +00:00
- [AI Sublime Text plugin ](https://github.com/yaroslavyaroslav/OpenAI-sublime-text ) (MIT)
2024-03-28 20:56:03 +00:00
*(to have a project listed here, it should clearly state that it depends on `llama.cpp` )*
2023-03-11 10:31:21 +00:00
---
2023-03-10 19:47:46 +00:00
2023-08-23 20:41:16 +00:00
Here is a typical run using LLaMA v2 13B on M2 Ultra:
2023-03-10 19:47:46 +00:00
2024-02-07 06:21:30 +00:00
```
2023-08-23 20:41:16 +00:00
$ make -j && ./main -m models/llama-13b-v2/ggml-model-q4_0.gguf -p "Building a website can be done in 10 simple steps:\nStep 1:" -n 400 -e
2023-08-23 20:43:00 +00:00
I llama.cpp build info:
2023-03-10 19:47:46 +00:00
I UNAME_S: Darwin
I UNAME_P: arm
I UNAME_M: arm64
2023-08-23 20:41:16 +00:00
I CFLAGS: -I. -O3 -std=c11 -fPIC -DNDEBUG -Wall -Wextra -Wpedantic -Wcast-qual -Wdouble-promotion -Wshadow -Wstrict-prototypes -Wpointer-arith -Wmissing-prototypes -pthread -DGGML_USE_K_QUANTS -DGGML_USE_ACCELERATE
I CXXFLAGS: -I. -I./common -O3 -std=c++11 -fPIC -DNDEBUG -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wno-multichar -pthread -DGGML_USE_K_QUANTS
2023-03-10 19:47:46 +00:00
I LDFLAGS: -framework Accelerate
2023-08-23 20:41:16 +00:00
I CC: Apple clang version 14.0.3 (clang-1403.0.22.14.1)
I CXX: Apple clang version 14.0.3 (clang-1403.0.22.14.1)
2023-03-10 19:47:46 +00:00
2023-03-10 22:09:19 +00:00
make: Nothing to be done for `default'.
2023-08-23 20:41:16 +00:00
main: build = 1041 (cf658ad)
main: seed = 1692823051
llama_model_loader: loaded meta data with 16 key-value pairs and 363 tensors from models/llama-13b-v2/ggml-model-q4_0.gguf (version GGUF V1 (latest))
llama_model_loader: - type f32: 81 tensors
llama_model_loader: - type q4_0: 281 tensors
llama_model_loader: - type q6_K: 1 tensors
llm_load_print_meta: format = GGUF V1 (latest)
llm_load_print_meta: arch = llama
llm_load_print_meta: vocab type = SPM
llm_load_print_meta: n_vocab = 32000
llm_load_print_meta: n_merges = 0
llm_load_print_meta: n_ctx_train = 4096
llm_load_print_meta: n_ctx = 512
llm_load_print_meta: n_embd = 5120
llm_load_print_meta: n_head = 40
llm_load_print_meta: n_head_kv = 40
llm_load_print_meta: n_layer = 40
llm_load_print_meta: n_rot = 128
llm_load_print_meta: n_gqa = 1
llm_load_print_meta: f_norm_eps = 1.0e-05
llm_load_print_meta: f_norm_rms_eps = 1.0e-05
llm_load_print_meta: n_ff = 13824
llm_load_print_meta: freq_base = 10000.0
llm_load_print_meta: freq_scale = 1
llm_load_print_meta: model type = 13B
llm_load_print_meta: model ftype = mostly Q4_0
llm_load_print_meta: model size = 13.02 B
llm_load_print_meta: general.name = LLaMA v2
llm_load_print_meta: BOS token = 1 '< s > '
llm_load_print_meta: EOS token = 2 '< / s > '
llm_load_print_meta: UNK token = 0 '< unk > '
llm_load_print_meta: LF token = 13 '< 0x0A > '
llm_load_tensors: ggml ctx size = 0.11 MB
llm_load_tensors: mem required = 7024.01 MB (+ 400.00 MB per state)
...................................................................................................
llama_new_context_with_model: kv self size = 400.00 MB
llama_new_context_with_model: compute buffer total size = 75.41 MB
2023-08-23 20:43:00 +00:00
system_info: n_threads = 16 / 24 | AVX = 0 | AVX2 = 0 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 0 | NEON = 1 | ARM_FMA = 1 | F16C = 0 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 0 | VSX = 0 |
2023-08-23 20:41:16 +00:00
sampling: repeat_last_n = 64, repeat_penalty = 1.100000, presence_penalty = 0.000000, frequency_penalty = 0.000000, top_k = 40, tfs_z = 1.000000, top_p = 0.950000, typical_p = 1.000000, temp = 0.800000, mirostat = 0, mirostat_lr = 0.100000, mirostat_ent = 5.000000
generate: n_ctx = 512, n_batch = 512, n_predict = 400, n_keep = 0
Building a website can be done in 10 simple steps:
Step 1: Find the right website platform.
Step 2: Choose your domain name and hosting plan.
Step 3: Design your website layout.
Step 4: Write your website content and add images.
Step 5: Install security features to protect your site from hackers or spammers
Step 6: Test your website on multiple browsers, mobile devices, operating systems etc…
Step 7: Test it again with people who are not related to you personally –
Step 8: Start marketing and promoting the website via social media channels or paid ads
Step 9: Analyze how many visitors have come to your site so far, what type of people visit more often than others (e.g., men vs women) etc…
Step 10: Continue to improve upon all aspects mentioned above by following trends in web design and staying up-to-date on new technologies that can enhance user experience even further!
How does a Website Work?
A website works by having pages, which are made of HTML code. This code tells your computer how to display the content on each page you visit – ’ ’
The most common type is called static HTML pages because they remain unchanged over time unless modified manually (either through editing files directly or using an interface such as WordPress). They are usually served up via HTTP protocols –
How to
llama_print_timings: load time = 576.45 ms
llama_print_timings: sample time = 283.10 ms / 400 runs ( 0.71 ms per token, 1412.91 tokens per second)
llama_print_timings: prompt eval time = 599.83 ms / 19 tokens ( 31.57 ms per token, 31.68 tokens per second)
llama_print_timings: eval time = 24513.59 ms / 399 runs ( 61.44 ms per token, 16.28 tokens per second)
llama_print_timings: total time = 25431.49 ms
2023-03-10 19:47:46 +00:00
```
2023-03-10 22:51:46 +00:00
And here is another demo of running both LLaMA-7B and [whisper.cpp ](https://github.com/ggerganov/whisper.cpp ) on a single M1 Pro MacBook:
https://user-images.githubusercontent.com/1991296/224442907-7693d4be-acaa-4e01-8b4f-add84093ffff.mp4
2023-03-10 19:47:46 +00:00
## Usage
2024-02-07 06:21:30 +00:00
Here are the end-to-end binary build and model conversion steps for most supported models.
2023-04-13 13:43:22 +00:00
### Get the Code
2023-03-11 08:47:09 +00:00
2023-03-10 19:47:46 +00:00
```bash
git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp
2023-04-13 13:43:22 +00:00
```
### Build
2023-04-26 20:03:03 +00:00
In order to build llama.cpp you have three different options.
2023-04-13 13:43:22 +00:00
2023-04-26 20:03:03 +00:00
- Using `make` :
- On Linux or MacOS:
2023-04-13 13:43:22 +00:00
2023-04-26 20:03:03 +00:00
```bash
make
```
2024-04-29 16:02:45 +00:00
**Note** : for `Debug` builds, run `make LLAMA_DEBUG=1`
2023-04-26 20:03:03 +00:00
- On Windows:
2023-04-28 14:22:48 +00:00
1. Download the latest fortran version of [w64devkit ](https://github.com/skeeto/w64devkit/releases ).
2023-04-26 20:03:03 +00:00
2. Extract `w64devkit` on your pc.
3. Run `w64devkit.exe` .
4. Use the `cd` command to reach the `llama.cpp` folder.
5. From here you can run:
```bash
make
```
2023-03-10 19:47:46 +00:00
2023-04-26 20:03:03 +00:00
- Using `CMake` :
2023-04-05 14:36:12 +00:00
2023-04-13 13:43:22 +00:00
```bash
2024-04-29 16:02:45 +00:00
cmake -B build
cmake --build build --config Release
2023-04-13 13:43:22 +00:00
```
2024-04-29 16:02:45 +00:00
**Note** : for `Debug` builds, there are two cases:
- Single-config generators (e.g. default = `Unix Makefiles` ; note that they just ignore the `--config` flag):
```bash
cmake -B build -DCMAKE_BUILD_TYPE=Debug
cmake --build build
```
- Multi-config generators (`-G` param set to Visual Studio, XCode...):
```bash
cmake -B build -G "Xcode"
cmake --build build --config Debug
```
2023-08-17 20:11:18 +00:00
- Using `Zig` (version 0.11 or later):
Building for optimization levels and CPU features can be accomplished using standard build arguments, for example AVX2, FMA, F16C,
it's also possible to cross compile for other operating systems and architectures:
2023-04-13 13:43:22 +00:00
```bash
2023-08-17 20:11:18 +00:00
zig build -Doptimize=ReleaseFast -Dtarget=x86_64-windows-gnu -Dcpu=x86_64+avx2+fma+f16c
2023-04-13 13:43:22 +00:00
```
2023-08-17 20:11:18 +00:00
The `zig targets` command will give you valid options to use.
2023-07-23 11:52:08 +00:00
- Using `gmake` (FreeBSD):
1. Install and activate [DRM in FreeBSD ](https://wiki.freebsd.org/Graphics )
2. Add your user to **video** group
3. Install compilation dependencies.
```bash
sudo pkg install gmake automake autoconf pkgconf llvm15 clinfo clover \
opencl clblast openblas
2024-01-30 09:16:38 +00:00
gmake CC=/usr/local/bin/clang15 CXX=/usr/local/bin/clang++15 -j4
2023-07-23 11:52:08 +00:00
```
**Notes:** With this packages you can build llama.cpp with OPENBLAS and
CLBLAST support for use OpenCL GPU acceleration in FreeBSD. Please read
the instructions for use and activate this options in this document below.
2023-06-04 20:34:30 +00:00
### Metal Build
2023-09-04 19:26:24 +00:00
On MacOS, Metal is enabled by default. Using Metal makes the computation run on the GPU.
To disable the Metal build at compile time use the `LLAMA_NO_METAL=1` flag or the `LLAMA_METAL=OFF` cmake option.
2023-06-04 20:34:30 +00:00
2023-10-12 11:10:50 +00:00
When built with Metal support, you can explicitly disable GPU inference with the `--n-gpu-layers|-ngl 0` command-line
2023-09-04 19:26:24 +00:00
argument.
2023-06-04 20:34:30 +00:00
2023-07-10 15:49:56 +00:00
### MPI Build
MPI lets you distribute the computation over a cluster of machines. Because of the serial nature of LLM prediction, this won't yield any end-to-end speed-ups, but it will let you run larger models than would otherwise fit into RAM on a single machine.
First you will need MPI libraries installed on your system. The two most popular (only?) options are [MPICH ](https://www.mpich.org ) and [OpenMPI ](https://www.open-mpi.org ). Either can be installed with a package manager (`apt`, Homebrew, MacPorts, etc).
Next you will need to build the project with `LLAMA_MPI` set to true on all machines; if you're building with `make` , you will also need to specify an MPI-capable compiler (when building with CMake, this is configured automatically):
- Using `make` :
```bash
make CC=mpicc CXX=mpicxx LLAMA_MPI=1
```
- Using `CMake` :
```bash
cmake -S . -B build -DLLAMA_MPI=ON
```
Once the programs are built, download/convert the weights on all of the machines in your cluster. The paths to the weights and programs should be identical on all machines.
Next, ensure password-less SSH access to each machine from the primary host, and create a `hostfile` with a list of the hostnames and their relative "weights" (slots). If you want to use localhost for computation, use its local subnet IP address rather than the loopback address or "localhost".
Here is an example hostfile:
```
192.168.0.1:2
malvolio.local:1
```
The above will distribute the computation across 2 processes on the first host and 1 process on the second host. Each process will use roughly an equal amount of RAM. Try to keep these numbers small, as inter-process (intra-host) communication is expensive.
Finally, you're ready to run a computation using `mpirun` :
```bash
2023-08-21 20:07:43 +00:00
mpirun -hostfile hostfile -n 3 ./main -m ./models/7B/ggml-model-q4_0.gguf -n 128
2023-07-10 15:49:56 +00:00
```
2023-04-26 20:03:03 +00:00
### BLAS Build
2023-11-30 21:49:21 +00:00
Building the program with BLAS support may lead to some performance improvements in prompt processing using batch sizes higher than 32 (the default is 512). Support with CPU-only BLAS implementations doesn't affect the normal generation performance. We may see generation performance improvements with GPU-involved BLAS implementations, e.g. cuBLAS, hipBLAS and CLBlast. There are currently several different BLAS implementations available for build and use:
2023-04-26 20:03:03 +00:00
2023-06-05 20:32:36 +00:00
- #### Accelerate Framework:
2023-04-26 20:03:03 +00:00
This is only available on Mac PCs and it's enabled by default. You can just build using the normal instructions.
2023-06-05 20:32:36 +00:00
- #### OpenBLAS:
2023-04-26 20:03:03 +00:00
This provides BLAS acceleration using only the CPU. Make sure to have OpenBLAS installed on your machine.
- Using `make` :
- On Linux:
```bash
make LLAMA_OPENBLAS=1
```
- On Windows:
1. Download the latest fortran version of [w64devkit ](https://github.com/skeeto/w64devkit/releases ).
2. Download the latest version of [OpenBLAS for Windows ](https://github.com/xianyi/OpenBLAS/releases ).
3. Extract `w64devkit` on your pc.
4. From the OpenBLAS zip that you just downloaded copy `libopenblas.a` , located inside the `lib` folder, inside `w64devkit\x86_64-w64-mingw32\lib` .
5. From the same OpenBLAS zip copy the content of the `include` folder inside `w64devkit\x86_64-w64-mingw32\include` .
6. Run `w64devkit.exe` .
7. Use the `cd` command to reach the `llama.cpp` folder.
8. From here you can run:
```bash
make LLAMA_OPENBLAS=1
```
- Using `CMake` on Linux:
```bash
2024-04-29 16:02:45 +00:00
cmake -B build -DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS
cmake --build build --config Release
2023-04-26 20:03:03 +00:00
```
2023-06-05 20:32:36 +00:00
- #### BLIS
2023-05-20 14:58:31 +00:00
2023-06-10 14:08:11 +00:00
Check [BLIS.md ](docs/BLIS.md ) for more information.
2023-05-20 14:58:31 +00:00
2024-02-02 07:56:31 +00:00
- #### SYCL
SYCL is a higher-level programming model to improve programming productivity on various hardware accelerators.
llama.cpp based on SYCL is used to **support Intel GPU** (Data Center Max series, Flex series, Arc series, Built-in GPU and iGPU).
For detailed info, please refer to [llama.cpp for SYCL ](README-sycl.md ).
2023-12-30 08:07:48 +00:00
- #### Intel oneMKL
2024-02-02 07:56:31 +00:00
Building through oneAPI compilers will make avx_vnni instruction set available for intel processors that do not support avx512 and avx512_vnni. Please note that this build config **does not support Intel GPU** . For Intel GPU support, please refer to [llama.cpp for SYCL ](./README-sycl.md ).
2023-12-30 08:07:48 +00:00
- Using manual oneAPI installation:
By default, `LLAMA_BLAS_VENDOR` is set to `Generic` , so if you already sourced intel environment script and assign `-DLLAMA_BLAS=ON` in cmake, the mkl version of Blas will automatically been selected. Otherwise please install oneAPI and follow the below steps:
```bash
2024-02-02 07:56:31 +00:00
source /opt/intel/oneapi/setvars.sh # You can skip this step if in oneapi-basekit docker image, only required for manual installation
2024-04-29 16:02:45 +00:00
cmake -B build -DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=Intel10_64lp -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DLLAMA_NATIVE=ON
cmake --build build --config Release
2023-12-30 08:07:48 +00:00
```
2023-05-20 14:58:31 +00:00
2023-12-30 08:07:48 +00:00
- Using oneAPI docker image:
2024-02-02 07:56:31 +00:00
If you do not want to source the environment vars and install oneAPI manually, you can also build the code using intel docker container: [oneAPI-basekit ](https://hub.docker.com/r/intel/oneapi-basekit ). Then, you can use the commands given above.
2023-12-30 08:07:48 +00:00
Check [Optimizing and Running LLaMA2 on Intel® CPU ](https://www.intel.com/content/www/us/en/content-details/791610/optimizing-and-running-llama2-on-intel-cpu.html ) for more information.
2023-05-20 14:58:31 +00:00
2024-03-26 00:16:01 +00:00
- #### CUDA
2023-04-26 20:03:03 +00:00
2024-03-26 00:16:01 +00:00
This provides GPU acceleration using the CUDA cores of your Nvidia GPU. Make sure to have the CUDA toolkit installed. You can download it from your Linux distro's package manager (e.g. `apt install nvidia-cuda-toolkit` ) or from here: [CUDA Toolkit ](https://developer.nvidia.com/cuda-downloads ).
2023-12-22 15:11:12 +00:00
For Jetson user, if you have Jetson Orin, you can try this: [Offical Support ](https://www.jetson-ai-lab.com/tutorial_text-generation.html ). If you are using an old model(nano/TX2), need some additional operations before compiling.
2023-04-26 20:03:03 +00:00
- Using `make` :
```bash
2024-03-26 00:16:01 +00:00
make LLAMA_CUDA=1
2023-04-26 20:03:03 +00:00
```
- Using `CMake` :
```bash
2024-04-29 16:02:45 +00:00
cmake -B build -DLLAMA_CUDA=ON
cmake --build build --config Release
2023-04-26 20:03:03 +00:00
```
2023-06-19 08:23:56 +00:00
The environment variable [`CUDA_VISIBLE_DEVICES` ](https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#env-vars ) can be used to specify which GPU(s) will be used. The following compilation options are also available to tweak performance:
2024-04-12 08:52:36 +00:00
| Option | Legal values | Default | Description |
|--------------------------------|------------------------|---------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| LLAMA_CUDA_FORCE_DMMV | Boolean | false | Force the use of dequantization + matrix vector multiplication kernels instead of using kernels that do matrix vector multiplication on quantized data. By default the decision is made based on compute capability (MMVQ for 6.1/Pascal/GTX 1000 or higher). Does not affect k-quants. |
| LLAMA_CUDA_DMMV_X | Positive integer >= 32 | 32 | Number of values in x direction processed by the CUDA dequantization + matrix vector multiplication kernel per iteration. Increasing this value can improve performance on fast GPUs. Power of 2 heavily recommended. Does not affect k-quants. |
| LLAMA_CUDA_MMV_Y | Positive integer | 1 | Block size in y direction for the CUDA mul mat vec kernels. Increasing this value can improve performance on fast GPUs. Power of 2 recommended. |
| LLAMA_CUDA_F16 | Boolean | false | If enabled, use half-precision floating point arithmetic for the CUDA dequantization + mul mat vec kernels and for the q4_1 and q5_1 matrix matrix multiplication kernels. Can improve performance on relatively recent GPUs. |
| LLAMA_CUDA_KQUANTS_ITER | 1 or 2 | 2 | Number of values processed per iteration and per CUDA thread for Q2_K and Q6_K quantization formats. Setting this value to 1 can improve performance for slow GPUs. |
| LLAMA_CUDA_PEER_MAX_BATCH_SIZE | Positive integer | 128 | Maximum batch size for which to enable peer access between multiple GPUs. Peer access requires either Linux or NVLink. When using NVLink enabling peer access for larger batch sizes is potentially beneficial. |
2023-06-03 13:35:20 +00:00
2023-08-25 09:09:42 +00:00
- #### hipBLAS
2023-09-08 16:04:32 +00:00
This provides BLAS acceleration on HIP-supported AMD GPUs.
2023-08-25 09:09:42 +00:00
Make sure to have ROCm installed.
2024-04-10 06:49:12 +00:00
You can download it from your Linux distro's package manager or from here: [ROCm Quick Start (Linux) ](https://rocm.docs.amd.com/projects/install-on-linux/en/latest/tutorial/quick-start.html#rocm-install-quick ).
2023-08-25 09:09:42 +00:00
- Using `make` :
```bash
make LLAMA_HIPBLAS=1
```
2023-12-21 19:45:32 +00:00
- Using `CMake` for Linux (assuming a gfx1030-compatible AMD GPU):
2023-08-25 09:09:42 +00:00
```bash
2023-12-21 19:45:32 +00:00
CC=/opt/rocm/llvm/bin/clang CXX=/opt/rocm/llvm/bin/clang++ \
2024-04-29 16:02:45 +00:00
cmake -B build -DLLAMA_HIPBLAS=ON -DAMDGPU_TARGETS=gfx1030 -DCMAKE_BUILD_TYPE=Release \
& & cmake --build build --config Release -- -j 16
2023-08-25 09:09:42 +00:00
```
2023-12-21 19:45:32 +00:00
On Linux it is also possible to use unified memory architecture (UMA) to share main memory between the CPU and integrated GPU by setting `-DLLAMA_HIP_UMA=ON"` .
2023-12-22 08:03:25 +00:00
However, this hurts performance for non-integrated GPUs (but enables working with integrated GPUs).
- Using `make` (example for target gfx1030, build with 16 CPU threads):
```bash
2024-04-09 22:23:02 +00:00
make -j16 LLAMA_HIPBLAS=1 LLAMA_HIP_UMA=1 AMDGPU_TARGETS=gfx1030
2023-12-22 08:03:25 +00:00
```
2023-12-21 19:45:32 +00:00
- Using `CMake` for Windows (using x64 Native Tools Command Prompt for VS, and assuming a gfx1100-compatible AMD GPU):
2023-11-20 15:02:46 +00:00
```bash
2023-11-24 07:52:39 +00:00
set PATH=%HIP_PATH%\bin;%PATH%
2023-11-20 15:02:46 +00:00
mkdir build
cd build
2024-04-07 18:52:19 +00:00
cmake -G Ninja -DAMDGPU_TARGETS=gfx1100 -DLLAMA_HIPBLAS=ON -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_BUILD_TYPE=Release ..
2023-11-20 15:02:46 +00:00
cmake --build .
```
Make sure that `AMDGPU_TARGETS` is set to the GPU arch you want to compile for. The above example uses `gfx1100` that corresponds to Radeon RX 7900XTX/XT/GRE. You can find a list of targets [here ](https://llvm.org/docs/AMDGPUUsage.html#processors )
2023-12-21 19:45:32 +00:00
Find your gpu version string by matching the most significant version information from `rocminfo | grep gfx | head -1 | awk '{print $2}'` with the list of processors, e.g. `gfx1035` maps to `gfx1030` .
2023-11-20 15:02:46 +00:00
2023-08-25 09:09:42 +00:00
The environment variable [`HIP_VISIBLE_DEVICES` ](https://rocm.docs.amd.com/en/latest/understand/gpu_isolation.html#hip-visible-devices ) can be used to specify which GPU(s) will be used.
2023-12-21 19:45:32 +00:00
If your GPU is not officially supported you can use the environment variable [`HSA_OVERRIDE_GFX_VERSION`] set to a similar GPU, for example 10.3.0 on RDNA2 (e.g. gfx1030, gfx1031, or gfx1035) or 11.0.0 on RDNA3.
2023-08-25 09:09:42 +00:00
The following compilation options are also available to tweak performance (yes, they refer to CUDA, not HIP, because it uses the same code as the cuBLAS version above):
2024-04-12 08:52:36 +00:00
| Option | Legal values | Default | Description |
|-------------------------|------------------------|---------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| LLAMA_CUDA_DMMV_X | Positive integer >= 32 | 32 | Number of values in x direction processed by the HIP dequantization + matrix vector multiplication kernel per iteration. Increasing this value can improve performance on fast GPUs. Power of 2 heavily recommended. Does not affect k-quants. |
| LLAMA_CUDA_MMV_Y | Positive integer | 1 | Block size in y direction for the HIP mul mat vec kernels. Increasing this value can improve performance on fast GPUs. Power of 2 recommended. Does not affect k-quants. |
| LLAMA_CUDA_KQUANTS_ITER | 1 or 2 | 2 | Number of values processed per iteration and per HIP thread for Q2_K and Q6_K quantization formats. Setting this value to 1 can improve performance for slow GPUs. |
2023-08-25 09:09:42 +00:00
2023-06-05 20:32:36 +00:00
- #### CLBlast
2023-05-27 15:47:55 +00:00
OpenCL acceleration is provided by the matrix multiplication kernels from the [CLBlast ](https://github.com/CNugteren/CLBlast ) project and custom kernels for ggml that can generate tokens on the GPU.
You will need the [OpenCL SDK ](https://github.com/KhronosGroup/OpenCL-SDK ).
2024-04-21 12:32:05 +00:00
- For Ubuntu, Debian, and Fedora the packages `opencl-headers` , `ocl-icd` may be needed.
2023-05-27 15:47:55 +00:00
2023-09-02 12:53:18 +00:00
- For Windows, a pre-built SDK is available on the [OpenCL Releases ](https://github.com/KhronosGroup/OpenCL-SDK/releases ) page.
2023-05-27 15:47:55 +00:00
- < details >
< summary > Installing the OpenCL SDK from source< / summary >
```sh
git clone --recurse-submodules https://github.com/KhronosGroup/OpenCL-SDK.git
2024-04-29 16:02:45 +00:00
cd OpenCL-SDK
cmake -B build -DBUILD_DOCS=OFF \
2023-05-27 15:47:55 +00:00
-DBUILD_EXAMPLES=OFF \
-DBUILD_TESTING=OFF \
-DOPENCL_SDK_BUILD_SAMPLES=OFF \
-DOPENCL_SDK_TEST_SAMPLES=OFF
2024-04-29 16:02:45 +00:00
cmake --build build
cmake --install build --prefix /some/path
2023-05-27 15:47:55 +00:00
```
< / details >
2023-09-02 12:53:18 +00:00
##### Installing CLBlast
Pre-built CLBlast binaries may be found on the [CLBlast Releases ](https://github.com/CNugteren/CLBlast/releases ) page. For Unix variants, it may also be found in your operating system's packages.
2024-04-21 12:32:05 +00:00
Linux packaging:
Fedora Linux:
```bash
sudo dnf install clblast
```
2023-09-02 12:53:18 +00:00
Alternatively, they may be built from source.
2023-05-27 15:47:55 +00:00
- < details >
2023-09-02 12:53:18 +00:00
< summary > Windows:< / summary >
```cmd
set OPENCL_SDK_ROOT="C:/OpenCL-SDK-v2023.04.17-Win-x64"
git clone https://github.com/CNugteren/CLBlast.git
2024-04-29 16:02:45 +00:00
cd CLBlast
cmake -B build -DBUILD_SHARED_LIBS=OFF -DOVERRIDE_MSVC_FLAGS_TO_MT=OFF -DTUNERS=OFF -DOPENCL_ROOT=%OPENCL_SDK_ROOT% -G "Visual Studio 17 2022" -A x64
cmake --build build --config Release
cmake --install build --prefix C:/CLBlast
2023-09-02 12:53:18 +00:00
```
2024-04-29 16:02:45 +00:00
(note: `--config Release` at build time is the default and only relevant for Visual Studio builds - or multi-config Ninja builds)
2023-09-02 12:53:18 +00:00
- < details >
< summary > Unix:< / summary >
2023-05-27 15:47:55 +00:00
```sh
git clone https://github.com/CNugteren/CLBlast.git
2024-04-29 16:02:45 +00:00
cd CLBlast
cmake -B build -DBUILD_SHARED_LIBS=OFF -DTUNERS=OFF
cmake --build build --config Release
cmake --install build --prefix /some/path
2023-05-27 15:47:55 +00:00
```
2023-06-03 13:35:20 +00:00
Where `/some/path` is where the built library will be installed (default is `/usr/local` ).
2023-05-27 15:47:55 +00:00
< / details >
2023-09-02 12:53:18 +00:00
##### Building Llama with CLBlast
2023-05-27 15:47:55 +00:00
- Build with make:
```sh
make LLAMA_CLBLAST=1
```
2023-09-02 12:53:18 +00:00
- CMake (Unix):
2023-05-27 15:47:55 +00:00
```sh
2024-04-29 16:02:45 +00:00
cmake -B build -DLLAMA_CLBLAST=ON -DCLBlast_DIR=/some/path
cmake --build build --config Release
2023-05-27 15:47:55 +00:00
```
2023-09-02 12:53:18 +00:00
- CMake (Windows):
```cmd
set CL_BLAST_CMAKE_PKG="C:/CLBlast/lib/cmake/CLBlast"
git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp
2024-04-29 16:02:45 +00:00
cmake -B build -DBUILD_SHARED_LIBS=OFF -DLLAMA_CLBLAST=ON -DCMAKE_PREFIX_PATH=%CL_BLAST_CMAKE_PKG% -G "Visual Studio 17 2022" -A x64
cmake --build build --config Release
cmake --install build --prefix C:/LlamaCPP
2023-09-02 12:53:18 +00:00
```
2023-05-27 15:47:55 +00:00
2023-09-02 12:53:18 +00:00
##### Running Llama with CLBlast
2023-05-27 15:47:55 +00:00
2023-06-04 20:34:30 +00:00
The CLBlast build supports `--gpu-layers|-ngl` like the CUDA version does.
2023-05-27 15:47:55 +00:00
To select the correct platform (driver) and device (GPU), you can use the environment variables `GGML_OPENCL_PLATFORM` and `GGML_OPENCL_DEVICE` .
The selection can be a number (starting from 0) or a text string to search:
```sh
GGML_OPENCL_PLATFORM=1 ./main ...
GGML_OPENCL_DEVICE=2 ./main ...
GGML_OPENCL_PLATFORM=Intel ./main ...
GGML_OPENCL_PLATFORM=AMD GGML_OPENCL_DEVICE=1 ./main ...
```
The default behavior is to find the first GPU device, but when it is an integrated GPU on a laptop, for instance, the selectors are useful.
Using the variables it is possible to select a CPU-based driver as well, if so desired.
You can get a list of platforms and devices from the `clinfo -l` command, etc.
2023-05-08 00:42:01 +00:00
2024-02-02 07:56:31 +00:00
- #### Vulkan
ggml : add unified SYCL backend for Intel GPUs (#2690)
* first update for migration
* update init_cublas
* add debug functio, commit all help code
* step 1
* step 2
* step3 add fp16, slower 31->28
* add GGML_LIST_DEVICE function
* step 5 format device and print
* step6, enhance error check, remove CUDA macro, enhance device id to fix none-zero id issue
* support main device is non-zero
* step7 add debug for code path, rm log
* step 8, rename all macro & func from cuda by sycl
* fix error of select non-zero device, format device list
* ren ggml-sycl.hpp -> ggml-sycl.h
* clear CMAKE to rm unused lib and options
* correct queue: rm dtct:get_queue
* add print tensor function to debug
* fix error: wrong result in 658746bb26702e50f2c59c0e4ada8e9da6010481
* summary dpct definition in one header file to replace folder:dpct
* refactor device log
* mv dpct definition from folder dpct to ggml-sycl.h
* update readme, refactor build script
* fix build with sycl
* set nthread=1 when sycl, increase performance
* add run script, comment debug code
* add ls-sycl-device tool
* add ls-sycl-device, rm unused files
* rm rear space
* dos2unix
* Update README_sycl.md
* fix return type
* remove sycl version from include path
* restore rm code to fix hang issue
* add syc and link for sycl readme
* rm original sycl code before refactor
* fix code err
* add know issue for pvc hang issue
* enable SYCL_F16 support
* align pr4766
* check for sycl blas, better performance
* cleanup 1
* remove extra endif
* add build&run script, clean CMakefile, update guide by review comments
* rename macro to intel hardware
* editor config format
* format fixes
* format fixes
* editor format fix
* Remove unused headers
* skip build sycl tool for other code path
* replace tab by space
* fix blas matmul function
* fix mac build
* restore hip dependency
* fix conflict
* ren as review comments
* mv internal function to .cpp file
* export funciton print_sycl_devices(), mv class dpct definition to source file
* update CI/action for sycl code, fix CI error of repeat/dup
* fix action ID format issue
* rm unused strategy
* enable llama_f16 in ci
* fix conflict
* fix build break on MacOS, due to CI of MacOS depend on external ggml, instead of internal ggml
* fix ci cases for unsupported data type
* revert unrelated changed in cuda cmake
remove useless nommq
fix typo of GGML_USE_CLBLAS_SYCL
* revert hip cmake changes
* fix indent
* add prefix in func name
* revert no mmq
* rm cpu blas duplicate
* fix no_new_line
* fix src1->type==F16 bug.
* pass batch offset for F16 src1
* fix batch error
* fix wrong code
* revert sycl checking in test-sampling
* pass void as arguments of ggml_backend_sycl_print_sycl_devices
* remove extra blank line in test-sampling
* revert setting n_threads in sycl
* implement std::isinf for icpx with fast math.
* Update ci/run.sh
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update examples/sycl/run-llama2.sh
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update examples/sycl/run-llama2.sh
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* add copyright and MIT license declare
* update the cmd example
---------
Co-authored-by: jianyuzh <jianyu.zhang@intel.com>
Co-authored-by: luoyu-intel <yu.luo@intel.com>
Co-authored-by: Meng, Hengyu <hengyu.meng@intel.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-01-28 15:56:23 +00:00
2024-02-02 07:56:31 +00:00
**With docker** :
ggml : add unified SYCL backend for Intel GPUs (#2690)
* first update for migration
* update init_cublas
* add debug functio, commit all help code
* step 1
* step 2
* step3 add fp16, slower 31->28
* add GGML_LIST_DEVICE function
* step 5 format device and print
* step6, enhance error check, remove CUDA macro, enhance device id to fix none-zero id issue
* support main device is non-zero
* step7 add debug for code path, rm log
* step 8, rename all macro & func from cuda by sycl
* fix error of select non-zero device, format device list
* ren ggml-sycl.hpp -> ggml-sycl.h
* clear CMAKE to rm unused lib and options
* correct queue: rm dtct:get_queue
* add print tensor function to debug
* fix error: wrong result in 658746bb26702e50f2c59c0e4ada8e9da6010481
* summary dpct definition in one header file to replace folder:dpct
* refactor device log
* mv dpct definition from folder dpct to ggml-sycl.h
* update readme, refactor build script
* fix build with sycl
* set nthread=1 when sycl, increase performance
* add run script, comment debug code
* add ls-sycl-device tool
* add ls-sycl-device, rm unused files
* rm rear space
* dos2unix
* Update README_sycl.md
* fix return type
* remove sycl version from include path
* restore rm code to fix hang issue
* add syc and link for sycl readme
* rm original sycl code before refactor
* fix code err
* add know issue for pvc hang issue
* enable SYCL_F16 support
* align pr4766
* check for sycl blas, better performance
* cleanup 1
* remove extra endif
* add build&run script, clean CMakefile, update guide by review comments
* rename macro to intel hardware
* editor config format
* format fixes
* format fixes
* editor format fix
* Remove unused headers
* skip build sycl tool for other code path
* replace tab by space
* fix blas matmul function
* fix mac build
* restore hip dependency
* fix conflict
* ren as review comments
* mv internal function to .cpp file
* export funciton print_sycl_devices(), mv class dpct definition to source file
* update CI/action for sycl code, fix CI error of repeat/dup
* fix action ID format issue
* rm unused strategy
* enable llama_f16 in ci
* fix conflict
* fix build break on MacOS, due to CI of MacOS depend on external ggml, instead of internal ggml
* fix ci cases for unsupported data type
* revert unrelated changed in cuda cmake
remove useless nommq
fix typo of GGML_USE_CLBLAS_SYCL
* revert hip cmake changes
* fix indent
* add prefix in func name
* revert no mmq
* rm cpu blas duplicate
* fix no_new_line
* fix src1->type==F16 bug.
* pass batch offset for F16 src1
* fix batch error
* fix wrong code
* revert sycl checking in test-sampling
* pass void as arguments of ggml_backend_sycl_print_sycl_devices
* remove extra blank line in test-sampling
* revert setting n_threads in sycl
* implement std::isinf for icpx with fast math.
* Update ci/run.sh
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update examples/sycl/run-llama2.sh
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update examples/sycl/run-llama2.sh
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* add copyright and MIT license declare
* update the cmd example
---------
Co-authored-by: jianyuzh <jianyu.zhang@intel.com>
Co-authored-by: luoyu-intel <yu.luo@intel.com>
Co-authored-by: Meng, Hengyu <hengyu.meng@intel.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-01-28 15:56:23 +00:00
2024-02-02 07:56:31 +00:00
You don't need to install Vulkan SDK. It will be installed inside the container.
ggml : add unified SYCL backend for Intel GPUs (#2690)
* first update for migration
* update init_cublas
* add debug functio, commit all help code
* step 1
* step 2
* step3 add fp16, slower 31->28
* add GGML_LIST_DEVICE function
* step 5 format device and print
* step6, enhance error check, remove CUDA macro, enhance device id to fix none-zero id issue
* support main device is non-zero
* step7 add debug for code path, rm log
* step 8, rename all macro & func from cuda by sycl
* fix error of select non-zero device, format device list
* ren ggml-sycl.hpp -> ggml-sycl.h
* clear CMAKE to rm unused lib and options
* correct queue: rm dtct:get_queue
* add print tensor function to debug
* fix error: wrong result in 658746bb26702e50f2c59c0e4ada8e9da6010481
* summary dpct definition in one header file to replace folder:dpct
* refactor device log
* mv dpct definition from folder dpct to ggml-sycl.h
* update readme, refactor build script
* fix build with sycl
* set nthread=1 when sycl, increase performance
* add run script, comment debug code
* add ls-sycl-device tool
* add ls-sycl-device, rm unused files
* rm rear space
* dos2unix
* Update README_sycl.md
* fix return type
* remove sycl version from include path
* restore rm code to fix hang issue
* add syc and link for sycl readme
* rm original sycl code before refactor
* fix code err
* add know issue for pvc hang issue
* enable SYCL_F16 support
* align pr4766
* check for sycl blas, better performance
* cleanup 1
* remove extra endif
* add build&run script, clean CMakefile, update guide by review comments
* rename macro to intel hardware
* editor config format
* format fixes
* format fixes
* editor format fix
* Remove unused headers
* skip build sycl tool for other code path
* replace tab by space
* fix blas matmul function
* fix mac build
* restore hip dependency
* fix conflict
* ren as review comments
* mv internal function to .cpp file
* export funciton print_sycl_devices(), mv class dpct definition to source file
* update CI/action for sycl code, fix CI error of repeat/dup
* fix action ID format issue
* rm unused strategy
* enable llama_f16 in ci
* fix conflict
* fix build break on MacOS, due to CI of MacOS depend on external ggml, instead of internal ggml
* fix ci cases for unsupported data type
* revert unrelated changed in cuda cmake
remove useless nommq
fix typo of GGML_USE_CLBLAS_SYCL
* revert hip cmake changes
* fix indent
* add prefix in func name
* revert no mmq
* rm cpu blas duplicate
* fix no_new_line
* fix src1->type==F16 bug.
* pass batch offset for F16 src1
* fix batch error
* fix wrong code
* revert sycl checking in test-sampling
* pass void as arguments of ggml_backend_sycl_print_sycl_devices
* remove extra blank line in test-sampling
* revert setting n_threads in sycl
* implement std::isinf for icpx with fast math.
* Update ci/run.sh
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update examples/sycl/run-llama2.sh
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update examples/sycl/run-llama2.sh
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* add copyright and MIT license declare
* update the cmd example
---------
Co-authored-by: jianyuzh <jianyu.zhang@intel.com>
Co-authored-by: luoyu-intel <yu.luo@intel.com>
Co-authored-by: Meng, Hengyu <hengyu.meng@intel.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-01-28 15:56:23 +00:00
2024-02-02 07:56:31 +00:00
```sh
# Build the image
docker build -t llama-cpp-vulkan -f .devops/main-vulkan.Dockerfile .
# Then, use it:
docker run -it --rm -v "$(pwd):/app:Z" --device /dev/dri/renderD128:/dev/dri/renderD128 --device /dev/dri/card1:/dev/dri/card1 llama-cpp-vulkan -m "/app/models/YOUR_MODEL_FILE" -p "Building a website can be done in 10 simple steps:" -n 400 -e -ngl 33
```
**Without docker** :
2024-02-07 06:21:30 +00:00
Firstly, you need to make sure you have installed [Vulkan SDK ](https://vulkan.lunarg.com/doc/view/latest/linux/getting_started_ubuntu.html )
ggml : add unified SYCL backend for Intel GPUs (#2690)
* first update for migration
* update init_cublas
* add debug functio, commit all help code
* step 1
* step 2
* step3 add fp16, slower 31->28
* add GGML_LIST_DEVICE function
* step 5 format device and print
* step6, enhance error check, remove CUDA macro, enhance device id to fix none-zero id issue
* support main device is non-zero
* step7 add debug for code path, rm log
* step 8, rename all macro & func from cuda by sycl
* fix error of select non-zero device, format device list
* ren ggml-sycl.hpp -> ggml-sycl.h
* clear CMAKE to rm unused lib and options
* correct queue: rm dtct:get_queue
* add print tensor function to debug
* fix error: wrong result in 658746bb26702e50f2c59c0e4ada8e9da6010481
* summary dpct definition in one header file to replace folder:dpct
* refactor device log
* mv dpct definition from folder dpct to ggml-sycl.h
* update readme, refactor build script
* fix build with sycl
* set nthread=1 when sycl, increase performance
* add run script, comment debug code
* add ls-sycl-device tool
* add ls-sycl-device, rm unused files
* rm rear space
* dos2unix
* Update README_sycl.md
* fix return type
* remove sycl version from include path
* restore rm code to fix hang issue
* add syc and link for sycl readme
* rm original sycl code before refactor
* fix code err
* add know issue for pvc hang issue
* enable SYCL_F16 support
* align pr4766
* check for sycl blas, better performance
* cleanup 1
* remove extra endif
* add build&run script, clean CMakefile, update guide by review comments
* rename macro to intel hardware
* editor config format
* format fixes
* format fixes
* editor format fix
* Remove unused headers
* skip build sycl tool for other code path
* replace tab by space
* fix blas matmul function
* fix mac build
* restore hip dependency
* fix conflict
* ren as review comments
* mv internal function to .cpp file
* export funciton print_sycl_devices(), mv class dpct definition to source file
* update CI/action for sycl code, fix CI error of repeat/dup
* fix action ID format issue
* rm unused strategy
* enable llama_f16 in ci
* fix conflict
* fix build break on MacOS, due to CI of MacOS depend on external ggml, instead of internal ggml
* fix ci cases for unsupported data type
* revert unrelated changed in cuda cmake
remove useless nommq
fix typo of GGML_USE_CLBLAS_SYCL
* revert hip cmake changes
* fix indent
* add prefix in func name
* revert no mmq
* rm cpu blas duplicate
* fix no_new_line
* fix src1->type==F16 bug.
* pass batch offset for F16 src1
* fix batch error
* fix wrong code
* revert sycl checking in test-sampling
* pass void as arguments of ggml_backend_sycl_print_sycl_devices
* remove extra blank line in test-sampling
* revert setting n_threads in sycl
* implement std::isinf for icpx with fast math.
* Update ci/run.sh
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update examples/sycl/run-llama2.sh
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update examples/sycl/run-llama2.sh
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* add copyright and MIT license declare
* update the cmd example
---------
Co-authored-by: jianyuzh <jianyu.zhang@intel.com>
Co-authored-by: luoyu-intel <yu.luo@intel.com>
Co-authored-by: Meng, Hengyu <hengyu.meng@intel.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-01-28 15:56:23 +00:00
2024-02-02 07:56:31 +00:00
For example, on Ubuntu 22.04 (jammy), use the command below:
```bash
wget -qO - https://packages.lunarg.com/lunarg-signing-key-pub.asc | apt-key add -
wget -qO /etc/apt/sources.list.d/lunarg-vulkan-jammy.list https://packages.lunarg.com/vulkan/lunarg-vulkan-jammy.list
apt update -y
apt-get install -y vulkan-sdk
# To verify the installation, use the command below:
vulkaninfo
```
2024-02-07 06:21:30 +00:00
Alternatively your package manager might be able to provide the appropiate libraries. For example for Ubuntu 22.04 you can install `libvulkan-dev` instead.
2024-02-02 07:56:31 +00:00
Then, build llama.cpp using the cmake command below:
```bash
2024-04-29 16:02:45 +00:00
cmake -B build -DLLAMA_VULKAN=1
cmake --build build --config Release
2024-02-02 07:56:31 +00:00
# Test the output binary (with "-ngl 33" to offload all layers to GPU)
./bin/main -m "PATH_TO_MODEL" -p "Hi you how are you" -n 50 -e -ngl 33 -t 4
# You should see in the output, ggml_vulkan detected your GPU. For example:
# ggml_vulkan: Using Intel(R) Graphics (ADL GT2) | uma: 1 | fp16: 1 | warp size: 32
```
ggml : add unified SYCL backend for Intel GPUs (#2690)
* first update for migration
* update init_cublas
* add debug functio, commit all help code
* step 1
* step 2
* step3 add fp16, slower 31->28
* add GGML_LIST_DEVICE function
* step 5 format device and print
* step6, enhance error check, remove CUDA macro, enhance device id to fix none-zero id issue
* support main device is non-zero
* step7 add debug for code path, rm log
* step 8, rename all macro & func from cuda by sycl
* fix error of select non-zero device, format device list
* ren ggml-sycl.hpp -> ggml-sycl.h
* clear CMAKE to rm unused lib and options
* correct queue: rm dtct:get_queue
* add print tensor function to debug
* fix error: wrong result in 658746bb26702e50f2c59c0e4ada8e9da6010481
* summary dpct definition in one header file to replace folder:dpct
* refactor device log
* mv dpct definition from folder dpct to ggml-sycl.h
* update readme, refactor build script
* fix build with sycl
* set nthread=1 when sycl, increase performance
* add run script, comment debug code
* add ls-sycl-device tool
* add ls-sycl-device, rm unused files
* rm rear space
* dos2unix
* Update README_sycl.md
* fix return type
* remove sycl version from include path
* restore rm code to fix hang issue
* add syc and link for sycl readme
* rm original sycl code before refactor
* fix code err
* add know issue for pvc hang issue
* enable SYCL_F16 support
* align pr4766
* check for sycl blas, better performance
* cleanup 1
* remove extra endif
* add build&run script, clean CMakefile, update guide by review comments
* rename macro to intel hardware
* editor config format
* format fixes
* format fixes
* editor format fix
* Remove unused headers
* skip build sycl tool for other code path
* replace tab by space
* fix blas matmul function
* fix mac build
* restore hip dependency
* fix conflict
* ren as review comments
* mv internal function to .cpp file
* export funciton print_sycl_devices(), mv class dpct definition to source file
* update CI/action for sycl code, fix CI error of repeat/dup
* fix action ID format issue
* rm unused strategy
* enable llama_f16 in ci
* fix conflict
* fix build break on MacOS, due to CI of MacOS depend on external ggml, instead of internal ggml
* fix ci cases for unsupported data type
* revert unrelated changed in cuda cmake
remove useless nommq
fix typo of GGML_USE_CLBLAS_SYCL
* revert hip cmake changes
* fix indent
* add prefix in func name
* revert no mmq
* rm cpu blas duplicate
* fix no_new_line
* fix src1->type==F16 bug.
* pass batch offset for F16 src1
* fix batch error
* fix wrong code
* revert sycl checking in test-sampling
* pass void as arguments of ggml_backend_sycl_print_sycl_devices
* remove extra blank line in test-sampling
* revert setting n_threads in sycl
* implement std::isinf for icpx with fast math.
* Update ci/run.sh
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update examples/sycl/run-llama2.sh
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update examples/sycl/run-llama2.sh
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update CMakeLists.txt
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* add copyright and MIT license declare
* update the cmd example
---------
Co-authored-by: jianyuzh <jianyu.zhang@intel.com>
Co-authored-by: luoyu-intel <yu.luo@intel.com>
Co-authored-by: Meng, Hengyu <hengyu.meng@intel.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-01-28 15:56:23 +00:00
2024-02-07 06:21:30 +00:00
### Prepare and Quantize
To obtain the official LLaMA 2 weights please see the < a href = "#obtaining-and-using-the-facebook-llama-2-model" > Obtaining and using the Facebook LLaMA 2 model</ a > section. There is also a large selection of pre-quantized `gguf` models available on Hugging Face.
2023-04-13 13:43:22 +00:00
2024-05-06 06:31:30 +00:00
Note: `convert.py` does not support LLaMA 3, you can use `convert-hf-to-gguf.py` with LLaMA 3 downloaded from Hugging Face.
2024-05-05 05:21:46 +00:00
2023-04-13 13:43:22 +00:00
```bash
2024-02-07 06:21:30 +00:00
# obtain the official LLaMA model weights and place them in ./models
2023-03-10 19:47:46 +00:00
ls ./models
2024-02-07 06:21:30 +00:00
llama-2-7b tokenizer_checklist.chk tokenizer.model
2024-01-30 09:16:38 +00:00
# [Optional] for models using BPE tokenizers
ls ./models
2024-02-07 06:21:30 +00:00
< folder containing weights and tokenizer json > vocab.json
# [Optional] for PyTorch .bin models like Mistral-7B
ls ./models
< folder containing weights and tokenizer json >
2023-03-10 19:47:46 +00:00
2023-03-11 05:47:26 +00:00
# install Python dependencies
py : new conversion script (#545)
Current status: Working, except for the latest GPTQ-for-LLaMa format
that includes `g_idx`. This turns out to require changes to GGML, so
for now it only works if you use the `--outtype` option to dequantize it
back to f16 (which is pointless except for debugging).
I also included some cleanup for the C++ code.
This script is meant to replace all the existing conversion scripts
(including the ones that convert from older GGML formats), while also
adding support for some new formats. Specifically, I've tested with:
- [x] `LLaMA` (original)
- [x] `llama-65b-4bit`
- [x] `alpaca-native`
- [x] `alpaca-native-4bit`
- [x] LLaMA converted to 'transformers' format using
`convert_llama_weights_to_hf.py`
- [x] `alpaca-native` quantized with `--true-sequential --act-order
--groupsize 128` (dequantized only)
- [x] same as above plus `--save_safetensors`
- [x] GPT4All
- [x] stock unversioned ggml
- [x] ggmh
There's enough overlap in the logic needed to handle these different
cases that it seemed best to move to a single script.
I haven't tried this with Alpaca-LoRA because I don't know where to find
it.
Useful features:
- Uses multiple threads for a speedup in some cases (though the Python
GIL limits the gain, and sometimes it's disk-bound anyway).
- Combines split models into a single file (both the intra-tensor split
of the original and the inter-tensor split of 'transformers' format
files). Single files are more convenient to work with and more
friendly to future changes to use memory mapping on the C++ side. To
accomplish this without increasing memory requirements, it has some
custom loading code which avoids loading whole input files into memory
at once.
- Because of the custom loading code, it no longer depends in PyTorch,
which might make installing dependencies slightly easier or faster...
although it still depends on NumPy and sentencepiece, so I don't know
if there's any meaningful difference. In any case, I also added a
requirements.txt file to lock the dependency versions in case of any
future breaking changes.
- Type annotations checked with mypy.
- Some attempts to be extra user-friendly:
- The script tries to be forgiving with arguments, e.g. you can
specify either the model file itself or the directory containing
it.
- The script doesn't depend on config.json / params.json, just in
case the user downloaded files individually and doesn't have those
handy. But you still need tokenizer.model and, for Alpaca,
added_tokens.json.
- The script tries to give a helpful error message if
added_tokens.json is missing.
2023-04-14 07:03:03 +00:00
python3 -m pip install -r requirements.txt
2023-03-11 05:47:26 +00:00
2024-02-07 06:21:30 +00:00
# convert the model to ggml FP16 format
python3 convert.py models/mymodel/
2023-03-10 19:47:46 +00:00
2024-01-30 09:16:38 +00:00
# [Optional] for models using BPE tokenizers
2024-02-07 21:11:30 +00:00
python convert.py models/mymodel/ --vocab-type bpe
2023-08-02 08:21:11 +00:00
2024-02-07 06:21:30 +00:00
# quantize the model to 4-bits (using Q4_K_M method)
./quantize ./models/mymodel/ggml-model-f16.gguf ./models/mymodel/ggml-model-Q4_K_M.gguf Q4_K_M
2023-03-10 19:47:46 +00:00
2024-02-07 06:21:30 +00:00
# update the gguf filetype to current version if older version is now unsupported
./quantize ./models/mymodel/ggml-model-Q4_K_M.gguf ./models/mymodel/ggml-model-Q4_K_M-v2.gguf COPY
```
2023-09-21 19:00:24 +00:00
2024-02-07 06:21:30 +00:00
### Run the quantized model
2023-09-21 19:00:24 +00:00
2024-02-07 06:21:30 +00:00
```bash
# start inference on a gguf model
./main -m ./models/mymodel/ggml-model-Q4_K_M.gguf -n 128
2023-03-10 19:47:46 +00:00
```
2023-03-11 08:47:09 +00:00
When running the larger models, make sure you have enough disk space to store all the intermediate files.
2023-10-17 18:13:21 +00:00
### Running on Windows with prebuilt binaries
You will find prebuilt Windows binaries on the release page.
Simply download and extract the latest zip package of choice: (e.g. `llama-b1380-bin-win-avx2-x64.zip` )
From the unzipped folder, open a terminal/cmd window here and place a pre-converted `.gguf` model file. Test out the main example like so:
```
.\main -m llama-2-7b.Q4_0.gguf -n 128
```
2023-03-18 20:58:46 +00:00
### Memory/Disk Requirements
2023-04-19 19:52:14 +00:00
As the models are currently fully loaded into memory, you will need adequate disk space to save them and sufficient RAM to load them. At the moment, memory and disk requirements are the same.
2023-03-18 20:58:46 +00:00
2024-02-07 06:21:30 +00:00
| Model | Original size | Quantized size (Q4_0) |
2024-04-12 08:52:36 +00:00
|------:|--------------:|----------------------:|
| 7B | 13 GB | 3.9 GB |
| 13B | 24 GB | 7.8 GB |
| 30B | 60 GB | 19.5 GB |
| 65B | 120 GB | 38.5 GB |
2023-03-13 16:15:20 +00:00
2023-04-26 20:24:42 +00:00
### Quantization
Several quantization methods are supported. They differ in the resulting model disk size and inference speed.
2023-08-23 20:41:16 +00:00
*(outdated)*
2024-04-12 08:52:36 +00:00
| Model | Measure | F16 | Q4_0 | Q4_1 | Q5_0 | Q5_1 | Q8_0 |
2023-05-11 21:23:08 +00:00
|------:|--------------|-------:|-------:|-------:|-------:|-------:|-------:|
2023-05-19 19:17:18 +00:00
| 7B | perplexity | 5.9066 | 6.1565 | 6.0912 | 5.9862 | 5.9481 | 5.9070 |
| 7B | file size | 13.0G | 3.5G | 3.9G | 4.3G | 4.7G | 6.7G |
| 7B | ms/tok @ 4th | 127 | 55 | 54 | 76 | 83 | 72 |
| 7B | ms/tok @ 8th | 122 | 43 | 45 | 52 | 56 | 67 |
| 7B | bits/weight | 16.0 | 4.5 | 5.0 | 5.5 | 6.0 | 8.5 |
| 13B | perplexity | 5.2543 | 5.3860 | 5.3608 | 5.2856 | 5.2706 | 5.2548 |
| 13B | file size | 25.0G | 6.8G | 7.6G | 8.3G | 9.1G | 13G |
| 13B | ms/tok @ 4th | - | 103 | 105 | 148 | 160 | 131 |
| 13B | ms/tok @ 8th | - | 73 | 82 | 98 | 105 | 128 |
| 13B | bits/weight | 16.0 | 4.5 | 5.0 | 5.5 | 6.0 | 8.5 |
2023-04-26 20:24:42 +00:00
2023-09-27 15:30:36 +00:00
- [k-quants ](https://github.com/ggerganov/llama.cpp/pull/1684 )
2024-02-06 17:00:16 +00:00
- recent k-quants improvements and new i-quants
2023-09-27 15:30:36 +00:00
- [#2707 ](https://github.com/ggerganov/llama.cpp/pull/2707 )
- [#2807 ](https://github.com/ggerganov/llama.cpp/pull/2807 )
2024-02-06 17:00:16 +00:00
- [#4773 - 2-bit i-quants (inference) ](https://github.com/ggerganov/llama.cpp/pull/4773 )
- [#4856 - 2-bit i-quants (inference) ](https://github.com/ggerganov/llama.cpp/pull/4856 )
- [#4861 - importance matrix ](https://github.com/ggerganov/llama.cpp/pull/4861 )
- [#4872 - MoE models ](https://github.com/ggerganov/llama.cpp/pull/4872 )
- [#4897 - 2-bit quantization ](https://github.com/ggerganov/llama.cpp/pull/4897 )
- [#4930 - imatrix for all k-quants ](https://github.com/ggerganov/llama.cpp/pull/4930 )
- [#4951 - imatrix on the GPU ](https://github.com/ggerganov/llama.cpp/pull/4957 )
- [#4969 - imatrix for legacy quants ](https://github.com/ggerganov/llama.cpp/pull/4969 )
- [#4996 - k-qunats tuning ](https://github.com/ggerganov/llama.cpp/pull/4996 )
- [#5060 - Q3_K_XS ](https://github.com/ggerganov/llama.cpp/pull/5060 )
- [#5196 - 3-bit i-quants ](https://github.com/ggerganov/llama.cpp/pull/5196 )
- [quantization tuning ](https://github.com/ggerganov/llama.cpp/pull/5320 ), [another one ](https://github.com/ggerganov/llama.cpp/pull/5334 ), and [another one ](https://github.com/ggerganov/llama.cpp/pull/5361 )
2023-09-27 15:30:36 +00:00
2023-05-08 14:41:54 +00:00
### Perplexity (measuring model quality)
You can use the `perplexity` example to measure perplexity over a given prompt (lower perplexity is better).
For more information, see [https://huggingface.co/docs/transformers/perplexity ](https://huggingface.co/docs/transformers/perplexity ).
The perplexity measurements in table above are done against the `wikitext2` test dataset (https://paperswithcode.com/dataset/wikitext-2), with context length of 512.
The time per token is measured on a MacBook M1 Pro 32GB RAM using 4 and 8 threads.
2023-10-06 19:13:36 +00:00
#### How to run
2024-02-18 20:39:30 +00:00
1. Download/extract: https://huggingface.co/datasets/ggml-org/ci/resolve/main/wikitext-2-raw-v1.zip
2023-10-06 19:13:36 +00:00
2. Run `./perplexity -m models/7B/ggml-model-q4_0.gguf -f wiki.test.raw`
3. Output:
```
perplexity : calculating perplexity over 655 chunks
24.43 seconds per pass - ETA 4.45 hours
[1]4.5970,[2]5.1807,[3]6.0382,...
```
And after 4.45 hours, you will have the final perplexity.
2023-03-12 21:13:28 +00:00
### Interactive mode
If you want a more ChatGPT-like experience, you can run in interactive mode by passing `-i` as a parameter.
2024-03-02 17:27:26 +00:00
In this mode, you can always interrupt generation by pressing Ctrl+C and entering one or more lines of text, which will be converted into tokens and appended to the current context. You can also specify a *reverse prompt* with the parameter `-r "reverse prompt string"` . This will result in user input being prompted whenever the exact tokens of the reverse prompt string are encountered in the generation. A typical use is to use a prompt that makes LLaMA emulate a chat between multiple users, say Alice and Bob, and pass `-r "Alice:"` .
2023-03-12 21:13:28 +00:00
2023-04-19 19:52:14 +00:00
Here is an example of a few-shot interaction, invoked with the command
2023-03-21 16:10:32 +00:00
```bash
2023-04-19 19:52:14 +00:00
# default arguments using a 7B model
2023-03-25 18:36:52 +00:00
./examples/chat.sh
2023-04-19 19:52:14 +00:00
# advanced chat with a 13B model
2023-03-25 18:36:52 +00:00
./examples/chat-13B.sh
2023-03-12 21:39:01 +00:00
2023-04-19 19:52:14 +00:00
# custom arguments using a 13B model
2023-08-21 20:07:43 +00:00
./main -m ./models/13B/ggml-model-q4_0.gguf -n 256 --repeat_penalty 1.0 --color -i -r "User:" -f prompts/chat-with-bob.txt
2023-03-12 21:13:28 +00:00
```
2023-03-21 16:10:32 +00:00
2023-04-24 15:45:32 +00:00
Note the use of `--color` to distinguish between user input and generated text. Other parameters are explained in more detail in the [README ](examples/main/README.md ) for the `main` example program.
2023-03-12 21:13:28 +00:00
2023-03-12 21:39:01 +00:00
2023-03-12 21:13:28 +00:00
2023-05-24 06:24:01 +00:00
### Persistent Interaction
The prompt, user inputs, and model generations can be saved and resumed across calls to `./main` by leveraging `--prompt-cache` and `--prompt-cache-all` . The `./examples/chat-persistent.sh` script demonstrates this with support for long-running, resumable chat sessions. To use this example, you must provide a file to cache the initial chat prompt and a directory to save the chat session, and may optionally provide the same variables as `chat-13B.sh` . The same prompt cache can be reused for new chat sessions. Note that both prompt cache and chat directory are tied to the initial prompt (`PROMPT_TEMPLATE`) and the model file.
```bash
# Start a new chat
PROMPT_CACHE_FILE=chat.prompt.bin CHAT_SAVE_DIR=./chat/default ./examples/chat-persistent.sh
# Resume that chat
PROMPT_CACHE_FILE=chat.prompt.bin CHAT_SAVE_DIR=./chat/default ./examples/chat-persistent.sh
# Start a different chat with the same prompt/model
PROMPT_CACHE_FILE=chat.prompt.bin CHAT_SAVE_DIR=./chat/another ./examples/chat-persistent.sh
# Different prompt cache for different prompt/model
PROMPT_TEMPLATE=./prompts/chat-with-bob.txt PROMPT_CACHE_FILE=bob.prompt.bin \
CHAT_SAVE_DIR=./chat/bob ./examples/chat-persistent.sh
```
2023-08-23 01:01:57 +00:00
### Constrained output with grammars
`llama.cpp` supports grammars to constrain model output. For example, you can force the model to output JSON only:
```bash
./main -m ./models/13B/ggml-model-q4_0.gguf -n 256 --grammar-file grammars/json.gbnf -p 'Request: schedule a call at 8pm; Command:'
```
The `grammars/` folder contains a handful of sample grammars. To write your own, check out the [GBNF Guide ](./grammars/README.md ).
2023-09-29 11:15:57 +00:00
For authoring more complex JSON grammars, you can also check out https://grammar.intrinsiclabs.ai/, a browser app that lets you write TypeScript interfaces which it compiles to GBNF grammars that you can save for local use. Note that the app is built and maintained by members of the community, please file any issues or FRs on [its repo ](http://github.com/intrinsiclabsai/gbnfgen ) and not this one.
2024-02-07 06:21:30 +00:00
### Instruct mode
2023-03-19 16:49:50 +00:00
2024-02-07 06:21:30 +00:00
1. First, download and place the `ggml` model into the `./models` folder
2023-03-23 11:30:40 +00:00
2. Run the `main` tool like this:
2023-03-19 16:49:50 +00:00
```
2023-03-25 18:36:52 +00:00
./examples/alpaca.sh
2023-03-19 16:49:50 +00:00
```
Sample run:
```
== Running in interactive mode. ==
- Press Ctrl+C to interject at any time.
2024-03-02 17:27:26 +00:00
- Press Return to return control to LLaMA.
2023-03-19 16:49:50 +00:00
- If you want to submit another line, end your input in '\'.
Below is an instruction that describes a task. Write a response that appropriately completes the request.
> How many letters are there in the English alphabet?
There 26 letters in the English Alphabet
> What is the most common way of transportation in Amsterdam?
The majority (54%) are using public transit. This includes buses, trams and metros with over 100 lines throughout the city which make it very accessible for tourists to navigate around town as well as locals who commute by tram or metro on a daily basis
2023-03-21 15:59:16 +00:00
> List 5 words that start with "ca".
2023-03-19 16:49:50 +00:00
cadaver, cauliflower, cabbage (vegetable), catalpa (tree) and Cailleach.
2023-04-11 19:45:44 +00:00
>
2023-03-19 16:49:50 +00:00
```
2023-07-28 01:14:11 +00:00
### Obtaining and using the Facebook LLaMA 2 model
- Refer to [Facebook's LLaMA download page ](https://ai.meta.com/resources/models-and-libraries/llama-downloads/ ) if you want to access the model data.
- Alternatively, if you want to save time and space, you can download already converted and quantized models from [TheBloke ](https://huggingface.co/TheBloke ), including:
2023-09-08 10:32:55 +00:00
- [LLaMA 2 7B base ](https://huggingface.co/TheBloke/Llama-2-7B-GGUF )
- [LLaMA 2 13B base ](https://huggingface.co/TheBloke/Llama-2-13B-GGUF )
- [LLaMA 2 70B base ](https://huggingface.co/TheBloke/Llama-2-70B-GGUF )
- [LLaMA 2 7B chat ](https://huggingface.co/TheBloke/Llama-2-7B-chat-GGUF )
- [LLaMA 2 13B chat ](https://huggingface.co/TheBloke/Llama-2-13B-chat-GGUF )
- [LLaMA 2 70B chat ](https://huggingface.co/TheBloke/Llama-2-70B-chat-GGUF )
2023-07-28 01:14:11 +00:00
2023-05-03 15:31:28 +00:00
### Seminal papers and background on the models
2023-03-20 20:14:06 +00:00
2023-05-03 15:31:28 +00:00
If your issue is with model generation quality, then please at least scan the following links and papers to understand the limitations of LLaMA models. This is especially important when choosing an appropriate model size and appreciating both the significant and subtle differences between LLaMA models and ChatGPT:
2023-04-19 19:52:14 +00:00
- LLaMA:
2023-05-03 15:31:28 +00:00
- [Introducing LLaMA: A foundational, 65-billion-parameter large language model ](https://ai.facebook.com/blog/large-language-model-llama-meta-ai/ )
- [LLaMA: Open and Efficient Foundation Language Models ](https://arxiv.org/abs/2302.13971 )
2023-04-19 19:52:14 +00:00
- GPT-3
2023-05-03 15:31:28 +00:00
- [Language Models are Few-Shot Learners ](https://arxiv.org/abs/2005.14165 )
2023-04-19 19:52:14 +00:00
- GPT-3.5 / InstructGPT / ChatGPT:
2023-05-03 15:31:28 +00:00
- [Aligning language models to follow instructions ](https://openai.com/research/instruction-following )
- [Training language models to follow instructions with human feedback ](https://arxiv.org/abs/2203.02155 )
2023-04-11 19:45:44 +00:00
2023-03-14 13:30:08 +00:00
### Android
2024-05-08 00:26:43 +00:00
#### Build on Android using Termux
[Termux ](https://github.com/termux/termux-app#installation ) is a method to execute `llama.cpp` on an Android device (no root required).
```
apt update & & apt upgrade -y
apt install git make cmake
```
2023-06-18 08:28:26 +00:00
2024-05-08 00:26:43 +00:00
It's recommended to move your model inside the `~/` directory for best performance:
2023-06-18 08:28:26 +00:00
```
2024-05-08 00:26:43 +00:00
cd storage/downloads
mv model.gguf ~/
2023-06-18 08:28:26 +00:00
```
2024-03-13 18:34:40 +00:00
2024-05-08 00:26:43 +00:00
[Get the code ](https://github.com/ggerganov/llama.cpp#get-the-code ) & [follow the Linux build instructions ](https://github.com/ggerganov/llama.cpp#build ) to build `llama.cpp` .
#### Building the Project using Android NDK
Obtain the [Android NDK ](https://developer.android.com/ndk ) and then build with CMake.
2024-03-13 18:34:40 +00:00
2024-05-08 00:26:43 +00:00
Execute the following commands on your computer to avoid downloading the NDK to your mobile. Alternatively, you can also do this in Termux:
2023-03-14 13:30:08 +00:00
```
$ mkdir build-android
$ cd build-android
$ export NDK=< your_ndk_directory >
$ cmake -DCMAKE_TOOLCHAIN_FILE=$NDK/build/cmake/android.toolchain.cmake -DANDROID_ABI=arm64-v8a -DANDROID_PLATFORM=android-23 -DCMAKE_C_FLAGS=-march=armv8.4a+dotprod ..
$ make
```
2024-05-08 00:26:43 +00:00
Install [termux ](https://github.com/termux/termux-app#installation ) on your device and run `termux-setup-storage` to get access to your SD card (if Android 11+ then run the command twice).
2024-03-13 18:34:40 +00:00
Finally, copy these built `llama` binaries and the model file to your device storage. Because the file permissions in the Android sdcard cannot be changed, you can copy the executable files to the `/data/data/com.termux/files/home/bin` path, and then execute the following commands in Termux to add executable permission:
(Assumed that you have pushed the built executable files to the /sdcard/llama.cpp/bin path using `adb push` )
```
$cp -r /sdcard/llama.cpp/bin /data/data/com.termux/files/home/
$cd /data/data/com.termux/files/home/bin
$chmod +x ./*
```
Download model [llama-2-7b-chat.Q4_K_M.gguf ](https://huggingface.co/TheBloke/Llama-2-7B-Chat-GGUF/blob/main/llama-2-7b-chat.Q4_K_M.gguf ), and push it to `/sdcard/llama.cpp/` , then move it to `/data/data/com.termux/files/home/model/`
```
$mv /sdcard/llama.cpp/llama-2-7b-chat.Q4_K_M.gguf /data/data/com.termux/files/home/model/
```
Now, you can start chatting:
```
$cd /data/data/com.termux/files/home/bin
$./main -m ../model/llama-2-7b-chat.Q4_K_M.gguf -n 128 -cml
```
2024-05-08 00:26:43 +00:00
Here's a demo of an interactive session running on Pixel 5 phone:
2023-03-14 13:30:08 +00:00
https://user-images.githubusercontent.com/271616/225014776-1d567049-ad71-4ef2-b050-55b0b3b9274c.mp4
2023-03-17 09:47:06 +00:00
### Docker
#### Prerequisites
* Docker must be installed and running on your system.
2023-04-19 19:52:14 +00:00
* Create a folder to store big models & intermediate files (ex. /llama/models)
2023-03-17 09:47:06 +00:00
#### Images
2024-01-28 07:55:31 +00:00
We have three Docker images available for this project:
2023-03-17 09:47:06 +00:00
2023-09-14 16:47:00 +00:00
1. `ghcr.io/ggerganov/llama.cpp:full` : This image includes both the main executable file and the tools to convert LLaMA models into ggml and convert into 4-bit quantization. (platforms: `linux/amd64` , `linux/arm64` )
2. `ghcr.io/ggerganov/llama.cpp:light` : This image only includes the main executable file. (platforms: `linux/amd64` , `linux/arm64` )
2024-02-14 15:15:49 +00:00
3. `ghcr.io/ggerganov/llama.cpp:server` : This image only includes the server executable file. (platforms: `linux/amd64` , `linux/arm64` )
2023-09-14 16:47:00 +00:00
Additionally, there the following images, similar to the above:
- `ghcr.io/ggerganov/llama.cpp:full-cuda` : Same as `full` but compiled with CUDA support. (platforms: `linux/amd64` )
- `ghcr.io/ggerganov/llama.cpp:light-cuda` : Same as `light` but compiled with CUDA support. (platforms: `linux/amd64` )
2024-01-28 07:55:31 +00:00
- `ghcr.io/ggerganov/llama.cpp:server-cuda` : Same as `server` but compiled with CUDA support. (platforms: `linux/amd64` )
2023-09-14 16:47:00 +00:00
- `ghcr.io/ggerganov/llama.cpp:full-rocm` : Same as `full` but compiled with ROCm support. (platforms: `linux/amd64` , `linux/arm64` )
- `ghcr.io/ggerganov/llama.cpp:light-rocm` : Same as `light` but compiled with ROCm support. (platforms: `linux/amd64` , `linux/arm64` )
2024-01-28 07:55:31 +00:00
- `ghcr.io/ggerganov/llama.cpp:server-rocm` : Same as `server` but compiled with ROCm support. (platforms: `linux/amd64` , `linux/arm64` )
2023-09-14 16:47:00 +00:00
2023-11-30 21:43:32 +00:00
The GPU enabled images are not currently tested by CI beyond being built. They are not built with any variation from the ones in the Dockerfiles defined in [.devops/ ](.devops/ ) and the GitHub Action defined in [.github/workflows/docker.yml ](.github/workflows/docker.yml ). If you need different settings (for example, a different CUDA or ROCm library, you'll need to build the images locally for now).
2023-03-17 09:47:06 +00:00
#### Usage
The easiest way to download the models, convert them to ggml and optimize them is with the --all-in-one command which includes the full docker image.
2023-04-06 06:56:58 +00:00
Replace `/path/to/models` below with the actual path where you downloaded the models.
2023-04-19 19:52:14 +00:00
```bash
2023-04-06 06:56:58 +00:00
docker run -v /path/to/models:/models ghcr.io/ggerganov/llama.cpp:full --all-in-one "/models/" 7B
2023-03-17 09:47:06 +00:00
```
2023-04-19 19:52:14 +00:00
On completion, you are ready to play!
2023-03-17 09:47:06 +00:00
```bash
2023-08-21 20:07:43 +00:00
docker run -v /path/to/models:/models ghcr.io/ggerganov/llama.cpp:full --run -m /models/7B/ggml-model-q4_0.gguf -p "Building a website can be done in 10 simple steps:" -n 512
2023-03-17 09:47:06 +00:00
```
2023-04-19 19:52:14 +00:00
or with a light image:
2023-03-17 09:47:06 +00:00
```bash
2023-08-21 20:07:43 +00:00
docker run -v /path/to/models:/models ghcr.io/ggerganov/llama.cpp:light -m /models/7B/ggml-model-q4_0.gguf -p "Building a website can be done in 10 simple steps:" -n 512
2023-03-17 09:47:06 +00:00
```
2023-03-14 13:30:08 +00:00
2024-01-28 07:55:31 +00:00
or with a server image:
```bash
docker run -v /path/to/models:/models -p 8000:8000 ghcr.io/ggerganov/llama.cpp:server -m /models/7B/ggml-model-q4_0.gguf --port 8000 --host 0.0.0.0 -n 512
```
2023-07-07 18:25:25 +00:00
### Docker With CUDA
Assuming one has the [nvidia-container-toolkit ](https://github.com/NVIDIA/nvidia-container-toolkit ) properly installed on Linux, or is using a GPU enabled cloud, `cuBLAS` should be accessible inside the container.
#### Building Locally
```bash
docker build -t local/llama.cpp:full-cuda -f .devops/full-cuda.Dockerfile .
docker build -t local/llama.cpp:light-cuda -f .devops/main-cuda.Dockerfile .
2024-01-28 07:55:31 +00:00
docker build -t local/llama.cpp:server-cuda -f .devops/server-cuda.Dockerfile .
2023-07-07 18:25:25 +00:00
```
You may want to pass in some different `ARGS` , depending on the CUDA environment supported by your container host, as well as the GPU architecture.
The defaults are:
- `CUDA_VERSION` set to `11.7.1`
- `CUDA_DOCKER_ARCH` set to `all`
The resulting images, are essentially the same as the non-CUDA images:
1. `local/llama.cpp:full-cuda` : This image includes both the main executable file and the tools to convert LLaMA models into ggml and convert into 4-bit quantization.
2. `local/llama.cpp:light-cuda` : This image only includes the main executable file.
2024-01-28 07:55:31 +00:00
3. `local/llama.cpp:server-cuda` : This image only includes the server executable file.
2023-07-07 18:25:25 +00:00
#### Usage
After building locally, Usage is similar to the non-CUDA examples, but you'll need to add the `--gpus` flag. You will also want to use the `--n-gpu-layers` flag.
```bash
2023-08-21 20:07:43 +00:00
docker run --gpus all -v /path/to/models:/models local/llama.cpp:full-cuda --run -m /models/7B/ggml-model-q4_0.gguf -p "Building a website can be done in 10 simple steps:" -n 512 --n-gpu-layers 1
docker run --gpus all -v /path/to/models:/models local/llama.cpp:light-cuda -m /models/7B/ggml-model-q4_0.gguf -p "Building a website can be done in 10 simple steps:" -n 512 --n-gpu-layers 1
2024-01-28 07:55:31 +00:00
docker run --gpus all -v /path/to/models:/models local/llama.cpp:server-cuda -m /models/7B/ggml-model-q4_0.gguf --port 8000 --host 0.0.0.0 -n 512 --n-gpu-layers 1
2023-07-07 18:25:25 +00:00
```
2023-03-13 07:42:26 +00:00
### Contributing
2023-03-13 17:21:51 +00:00
- Contributors can open PRs
2023-03-16 06:55:13 +00:00
- Collaborators can push to branches in the `llama.cpp` repo and merge PRs into the `master` branch
2023-03-13 07:42:26 +00:00
- Collaborators will be invited based on contributions
2023-03-16 06:55:13 +00:00
- Any help with managing issues and PRs is very appreciated!
2023-03-17 18:30:04 +00:00
- Make sure to read this: [Inference at the edge ](https://github.com/ggerganov/llama.cpp/discussions/205 )
2023-03-23 08:46:58 +00:00
- A bit of backstory for those who are interested: [Changelog podcast ](https://changelog.com/podcast/532 )
2023-03-13 07:42:26 +00:00
2023-03-14 07:43:52 +00:00
### Coding guidelines
2023-03-13 07:42:26 +00:00
- Avoid adding third-party dependencies, extra files, extra headers, etc.
- Always consider cross-compatibility with other operating systems and architectures
2023-03-14 07:43:52 +00:00
- Avoid fancy looking modern STL constructs, use basic `for` loops, avoid templates, keep it simple
2023-03-13 07:42:26 +00:00
- There are no strict rules for the code style, but try to follow the patterns in the code (indentation, spaces, etc.). Vertical alignment makes things more readable and easier to batch edit
2023-04-19 19:52:14 +00:00
- Clean-up any trailing whitespaces, use 4 spaces for indentation, brackets on the same line, `void * ptr` , `int & a`
2023-03-14 07:43:52 +00:00
- See [good first issues ](https://github.com/ggerganov/llama.cpp/issues?q=is%3Aissue+is%3Aopen+label%3A%22good+first+issue%22 ) for tasks suitable for first contributions
2023-12-21 17:27:14 +00:00
- Tensors store data in row-major order. We refer to dimension 0 as columns, 1 as rows, 2 as matrices
2024-04-24 19:29:13 +00:00
- Matrix multiplication is unconventional: [`C = ggml_mul_mat(ctx, A, B)` ](https://github.com/ggerganov/llama.cpp/blob/880e352277fc017df4d5794f0c21c44e1eae2b84/ggml.h#L1058-L1064 ) means $C^T = A B^T \Leftrightarrow C = B A^T.$
2023-03-23 11:30:40 +00:00
2023-04-05 15:56:20 +00:00
### Docs
2023-07-09 07:38:42 +00:00
- [main ](./examples/main/README.md )
- [server ](./examples/server/README.md )
- [jeopardy ](./examples/jeopardy/README.md )
- [BLIS ](./docs/BLIS.md )
2023-06-05 20:32:36 +00:00
- [Performance troubleshooting ](./docs/token_generation_performance_tips.md )
2023-07-09 07:38:42 +00:00
- [GGML tips & tricks ](https://github.com/ggerganov/llama.cpp/wiki/GGML-Tips-&-Tricks )
2023-08-23 01:01:57 +00:00
- [GBNF grammars ](./grammars/README.md )
