3859 Commits

This Branch

This Branch

All Branches

Author	SHA1	Message	Date
ochafik	9cfe4d7202	tool-call: refactor llama_chat_template class + use in validate_model_chat_template	2024-09-26 18:06:03 +01:00
ochafik	cf7bece6a7	tool-call: factor chat template away from legacy API	2024-09-26 17:19:29 +01:00
ochafik	d7ec84f78c	tool-call: allow <|python_tag|> in functionary-medium-3.1	2024-09-26 06:52:34 +01:00
ochafik	3d2650ce65	fix gcc build	2024-09-26 06:52:34 +01:00
ochafik	749a21c67a	gcc appeasement	2024-09-26 06:08:18 +01:00
ochafik	0c870133d8	tool-call: test/fix functionary-medium-v3.1's template (can "look" like llama3.1 template)	2024-09-26 05:56:15 +01:00
ochafik	8e4a9bad8a	minja: allow none input to selectattr, and add safe passthrough filter	2024-09-26 05:53:12 +01:00
ochafik	5f5be9cde7	minja: gcc tweaks	2024-09-26 05:06:11 +01:00
ochafik	2eb29bf8b8	tool-call: update chat templates/goldens	2024-09-26 04:00:10 +01:00
ochafik	4cd82d61dd	tool-call: fix pyright type errors	2024-09-26 03:59:38 +01:00
ochafik	059babdd9b	minja: try to please gcc	2024-09-26 03:58:18 +01:00
ochafik	94377d743c	server: catch errors in format_final_response_oaicompat instead of taking server down	2024-09-26 03:42:36 +01:00
ochafik	595e11cb11	tool-call: fix/test functionary v3	2024-09-26 03:42:05 +01:00
ochafik	c124ab48ea	minja: add str.endswith	2024-09-26 03:21:23 +01:00
ochafik	76d2938ef8	fix flake8 lints	2024-09-26 02:30:17 +01:00
ochafik	1b6280102b	fix editorconfig lints	2024-09-26 02:27:46 +01:00
ochafik	ab25e3fbf9	tool-call: allow empty message content when there's tool_calls in format_chat	2024-09-26 02:19:04 +01:00
ochafik	d928ff4dfd	server: catch errors in oaicompat_completion_params_parse instead of taking server down	2024-09-26 02:18:01 +01:00
ochafik	a774093a99	tool-call: add server tests for llama 3.1	2024-09-26 02:17:30 +01:00
ochafik	9e366b3d03	server: fix tailing comma in completions_seed	2024-09-26 02:15:48 +01:00
ochafik	45b243b4a5	minja: fix llama_chat_apply_template + adde use_jinja param to validate_model_chat_template	2024-09-26 02:14:42 +01:00
ochafik	e983c9d0de	tool-call: fix llama_chat_apply_template signature / test-chat-template	2024-09-25 22:02:58 +01:00
ochafik	97d0620968	minja: fetch more templates (add models from test-chat-template)	2024-09-25 19:22:43 +01:00
ochafik	d15dcfb09d	tool-call: add output example to readme	2024-09-25 19:22:16 +01:00
ochafik	33ea20edd1	Merge remote-tracking branch 'origin/master' into tool-call	2024-09-25 18:58:54 +01:00
ochafik	8f25531c44	tool-call: add basic usage example to server readme	2024-09-25 18:00:31 +01:00
ochafik	4706bdbae1	tool-call: support Functionary v3 vs. v3-llama3.1 variants	2024-09-25 17:33:00 +01:00
Xuan Son Nguyen	ea9c32be71	ci : fix docker build number and tag name (#9638) ... * ci : fix docker build number and tag name * fine-grant permissions	2024-09-25 17:26:01 +02:00
ochafik	41103c0ed6	server: add --chat-template-file	2024-09-25 16:14:46 +01:00
ochafik	e309c6a47f	tool-call: integrate minja & tool-call to server when --jinja is set	2024-09-25 16:14:46 +01:00
ochafik	3cfc21ea71	tool-call: basic Functionary 3.2, Llama 3.1, Hermes 2 Pro grammar generators + parsers	2024-09-25 16:14:22 +01:00
ochafik	26c175b416	json: build_grammar helper	2024-09-25 16:14:22 +01:00
ochafik	eaca756ecc	minja: minimalist Jinja templating engine for LLM chat templates	2024-09-25 16:14:22 +01:00
ochafik	5b6d5040d5	grammar: trigger words + refactor of antiprompts	2024-09-25 16:14:22 +01:00
Charles Xu	1e43630218	ggml : remove assert for AArch64 GEMV and GEMM Q4 kernels (#9217) ... * ggml : remove assert for AArch64 GEMV and GEMM Q4 kernels * added fallback mechanism when the offline re-quantized model is not optimized for the underlying target. * fix for build errors * remove prints from the low-level code * Rebase to the latest upstream	2024-09-25 16:12:20 +03:00
Xuan Son Nguyen	afbbfaa537	server : add more env vars, improve gen-docs (#9635) ... * server : add more env vars, improve gen-docs * update server docs * LLAMA_ARG_NO_CONTEXT_SHIFT	2024-09-25 14:05:13 +02:00
Gabe Goodhart	3d6bf6919f	llama : add IBM Granite MoE architecture (#9438) ... * feat(gguf-py): Add granitemoe architecture This includes the addition of new tensor names for the new moe layers. These may not be correct at this point due to the need for the hack in gguf_writer.py to double-check the length of the shape for these layers. Branch: GraniteMoE Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat(convert_hf_to_gguf): Add GraniteMoeModel GraniteMoe has the same configuration deltas as Granite Branch: GraniteMoE Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(granitemoe convert): Split the double-sized input layer into gate and up After a lot of staring and squinting, it's clear that the standard mixtral expert implementation is equivalent to the vectorized parallel experts in granite. The difference is that in granite, the w1 and w3 are concatenated into a single tensor "input_linear." Rather than reimplementing all of the math on the llama.cpp side, the much simpler route is to just split this tensor during conversion and follow the standard mixtral route. Branch: GraniteMoE Co-Authored-By: alex.brooks@ibm.com Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat(granitemoe): Implement granitemoe GraniteMoE follows the mixtral architecture (once the input_linear layers are split into gate_exps/up_exps). The main delta is the addition of the same four multipliers used in Granite. Branch: GraniteMoE Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * Typo fix in docstring Co-Authored-By: ggerganov@gmail.com Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(conversion): Simplify tensor name mapping in conversion Branch: GraniteMoE Co-Authored-By: git@compilade.net Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(convert): Remove unused tensor name mappings Branch: GraniteMoE Co-Authored-By: git@compilade.net Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(convert): Sanity check on merged FFN tensor sizes Branch: GraniteMoE Co-Authored-By: git@compilade.net Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Allow "output" layer in granite moe architecture (convert and cpp) Branch: GraniteMoE Co-Authored-By: git@compilade.net Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(granite): Add missing 'output' tensor for Granite This is a fix for the previous `granite` architecture PR. Recent snapshots have included this (`lm_head.weights`) as part of the architecture Branch: GraniteMoE Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> --------- Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>	2024-09-25 10:06:52 +03:00
Dou Xinpeng	904837e0cb	cann: fix crash when llama-bench is running on multiple cann devices (#9627)	2024-09-25 11:30:38 +08:00
Eric Zhang	70392f1f81	ggml : add AVX512DQ requirement for AVX512 builds (#9622)	2024-09-24 11:03:21 +03:00
Georgi Gerganov	bb5f819975	sync : ggml	2024-09-24 11:01:18 +03:00
Georgi Gerganov	c038931615	examples : adapt to ggml.h changes (ggml/0) ... ggml-ci	2024-09-24 11:00:52 +03:00
Georgi Gerganov	31ac5834fe	llama : keep track of all EOG tokens in the vocab (#9609) ... ggml-ci	2024-09-24 10:16:06 +03:00
Georgi Gerganov	cea1486ecf	log : add CONT level for continuing previous log entry (#9610)	2024-09-24 10:15:35 +03:00
StrangeBytesDev	0aa15011e3	server : add newline after chat example (#9616)	2024-09-24 09:04:39 +03:00
Georgi Gerganov	b0f27361f3	sampling : avoid expensive softmax during greedy sampling (#9605) ... * sampling : avoid expensive softmax during greedy sampling ggml-ci * speculative : fix default RNG seed + set sparams.n_probs * Update tests/test-sampling.cpp Co-authored-by: slaren <slarengh@gmail.com> * sampling : add clarifying comment [no ci] --------- Co-authored-by: slaren <slarengh@gmail.com>	2024-09-24 09:03:17 +03:00
Max Krasnyansky	c087b6f11d	threads: fix msvc build without openmp (#9615) ... We're missing atomic_thread_fence() in MSVC builds when openmp is disabled.	2024-09-23 21:18:48 -07:00
Ivan	116efee0ee	cuda: add q8_0->f32 cpy operation (#9571) ... llama: enable K-shift for quantized KV cache It will fail on unsupported backends or quant types.	2024-09-24 02:14:24 +02:00
Xuan Son Nguyen	0b3bf966f4	server : add --no-context-shift option (#9607) ... * server : add --no-context-shift option * small fix * Update examples/server/tests/features/embeddings.feature Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> * tests : minor fix * revert usage of GGML_ASSERT * update server documentation --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>	2024-09-23 22:23:54 +02:00
Max Krasnyansky	f0c7b5edf8	threads: improve ggml_barrier scaling with large number of threads (#9598) ... Make sure n_barrier and n_barrier_passed do not share the cache line to avoid cache line bouncing. This optimization shows performance improvements even for n_threads <= 8 cases. Resurect TSAN (Thread Sanitizer) check so that we can avoid doing expensive read-modify-write in the normal case and just use thread-fence as originally intended. --- Here is the original description and suggestions from Willy Tarreau : There's currently some false sharing between n_barrier and n_barrier_passed that is amplified in ggml_barrier() by the fact that all threads need to increment n_barrier when entering, while all previous threads continue to read n_barrier_passed, waiting for the last one to release them all. The side effect is that all these readers are slowing down all new threads by making the cache line bounce back and forth between readers and writers. Just placing them in two distinct cache lines is sufficient to boost the performance by 21% on a 80-core ARM server compared to the no-openmp version, and by 3% compared to the openmp version. Note that the variables could have been spread apart in the structure as well, but it doesn't seem that the size of this threadpool struct is critical so here we're simply aligning them. Finally, the same issue was present when leaving the barrier since all threads had to update the n_barrier_passed counter, though only one would add a non-zero value. This alone is responsible for half of the cost due to undesired serialization. It might be possible that using a small array of n_barrier counters could make things even faster on many-core systems, but it would likely complicate the logic needed to detect the last thread. Co-authored-by: Willy Tarreau <w@1wt.eu>	2024-09-23 11:42:43 -07:00
Riceball LEE	1d48e98e4f	readme : add programmable prompt engine language CLI (#9599)	2024-09-23 18:58:17 +03:00