mirror of
https://github.com/ggerganov/llama.cpp.git
synced 2024-12-24 10:24:35 +00:00
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
This commit is contained in:
parent
3d6bf6919f
commit
afbbfaa537
@ -691,7 +691,7 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
|
||||
[](gpt_params & params) {
|
||||
params.ctx_shift = false;
|
||||
}
|
||||
).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER}));
|
||||
).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_NO_CONTEXT_SHIFT"));
|
||||
add_opt(llama_arg(
|
||||
{"--chunks"}, "N",
|
||||
format("max number of chunks to process (default: %d, -1 = all)", params.n_chunks),
|
||||
@ -1102,7 +1102,7 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
|
||||
else if (value == "last") { params.pooling_type = LLAMA_POOLING_TYPE_LAST; }
|
||||
else { throw std::invalid_argument("invalid value"); }
|
||||
}
|
||||
).set_examples({LLAMA_EXAMPLE_EMBEDDING}));
|
||||
).set_examples({LLAMA_EXAMPLE_EMBEDDING, LLAMA_EXAMPLE_RETRIEVAL, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_POOLING"));
|
||||
add_opt(llama_arg(
|
||||
{"--attention"}, "{causal,non,causal}",
|
||||
"attention type for embeddings, use model default if unspecified",
|
||||
@ -1121,77 +1121,77 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
|
||||
else if (value == "yarn") { params.rope_scaling_type = LLAMA_ROPE_SCALING_TYPE_YARN; }
|
||||
else { throw std::invalid_argument("invalid value"); }
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_ROPE_SCALING_TYPE"));
|
||||
add_opt(llama_arg(
|
||||
{"--rope-scale"}, "N",
|
||||
"RoPE context scaling factor, expands context by a factor of N",
|
||||
[](gpt_params & params, const std::string & value) {
|
||||
params.rope_freq_scale = 1.0f / std::stof(value);
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_ROPE_SCALE"));
|
||||
add_opt(llama_arg(
|
||||
{"--rope-freq-base"}, "N",
|
||||
"RoPE base frequency, used by NTK-aware scaling (default: loaded from model)",
|
||||
[](gpt_params & params, const std::string & value) {
|
||||
params.rope_freq_base = std::stof(value);
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_ROPE_FREQ_BASE"));
|
||||
add_opt(llama_arg(
|
||||
{"--rope-freq-scale"}, "N",
|
||||
"RoPE frequency scaling factor, expands context by a factor of 1/N",
|
||||
[](gpt_params & params, const std::string & value) {
|
||||
params.rope_freq_scale = std::stof(value);
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_ROPE_FREQ_SCALE"));
|
||||
add_opt(llama_arg(
|
||||
{"--yarn-orig-ctx"}, "N",
|
||||
format("YaRN: original context size of model (default: %d = model training context size)", params.yarn_orig_ctx),
|
||||
[](gpt_params & params, int value) {
|
||||
params.yarn_orig_ctx = value;
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_YARN_ORIG_CTX"));
|
||||
add_opt(llama_arg(
|
||||
{"--yarn-ext-factor"}, "N",
|
||||
format("YaRN: extrapolation mix factor (default: %.1f, 0.0 = full interpolation)", (double)params.yarn_ext_factor),
|
||||
[](gpt_params & params, const std::string & value) {
|
||||
params.yarn_ext_factor = std::stof(value);
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_YARN_EXT_FACTOR"));
|
||||
add_opt(llama_arg(
|
||||
{"--yarn-attn-factor"}, "N",
|
||||
format("YaRN: scale sqrt(t) or attention magnitude (default: %.1f)", (double)params.yarn_attn_factor),
|
||||
[](gpt_params & params, const std::string & value) {
|
||||
params.yarn_attn_factor = std::stof(value);
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_YARN_ATTN_FACTOR"));
|
||||
add_opt(llama_arg(
|
||||
{"--yarn-beta-slow"}, "N",
|
||||
format("YaRN: high correction dim or alpha (default: %.1f)", (double)params.yarn_beta_slow),
|
||||
[](gpt_params & params, const std::string & value) {
|
||||
params.yarn_beta_slow = std::stof(value);
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_YARN_BETA_SLOW"));
|
||||
add_opt(llama_arg(
|
||||
{"--yarn-beta-fast"}, "N",
|
||||
format("YaRN: low correction dim or beta (default: %.1f)", (double)params.yarn_beta_fast),
|
||||
[](gpt_params & params, const std::string & value) {
|
||||
params.yarn_beta_fast = std::stof(value);
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_YARN_BETA_FAST"));
|
||||
add_opt(llama_arg(
|
||||
{"-gan", "--grp-attn-n"}, "N",
|
||||
format("group-attention factor (default: %d)", params.grp_attn_n),
|
||||
[](gpt_params & params, int value) {
|
||||
params.grp_attn_n = value;
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_GRP_ATTN_N"));
|
||||
add_opt(llama_arg(
|
||||
{"-gaw", "--grp-attn-w"}, "N",
|
||||
format("group-attention width (default: %.1f)", (double)params.grp_attn_w),
|
||||
[](gpt_params & params, int value) {
|
||||
params.grp_attn_w = value;
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_GRP_ATTN_W"));
|
||||
add_opt(llama_arg(
|
||||
{"-dkvc", "--dump-kv-cache"},
|
||||
"verbose print of the KV cache",
|
||||
@ -1205,7 +1205,7 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
|
||||
[](gpt_params & params) {
|
||||
params.no_kv_offload = true;
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_NO_KV_OFFLOAD"));
|
||||
add_opt(llama_arg(
|
||||
{"-ctk", "--cache-type-k"}, "TYPE",
|
||||
format("KV cache data type for K (default: %s)", params.cache_type_k.c_str()),
|
||||
@ -1213,7 +1213,7 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
|
||||
// TODO: get the type right here
|
||||
params.cache_type_k = value;
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_CACHE_TYPE_K"));
|
||||
add_opt(llama_arg(
|
||||
{"-ctv", "--cache-type-v"}, "TYPE",
|
||||
format("KV cache data type for V (default: %s)", params.cache_type_v.c_str()),
|
||||
@ -1221,7 +1221,7 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
|
||||
// TODO: get the type right here
|
||||
params.cache_type_v = value;
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_CACHE_TYPE_V"));
|
||||
add_opt(llama_arg(
|
||||
{"--perplexity", "--all-logits"},
|
||||
format("return logits for all tokens in the batch (default: %s)", params.logits_all ? "true" : "false"),
|
||||
@ -1355,7 +1355,7 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
|
||||
[](gpt_params & params, const std::string & value) {
|
||||
params.rpc_servers = value;
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_RPC"));
|
||||
#endif
|
||||
add_opt(llama_arg(
|
||||
{"--mlock"},
|
||||
@ -1363,14 +1363,14 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
|
||||
[](gpt_params & params) {
|
||||
params.use_mlock = true;
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_MLOCK"));
|
||||
add_opt(llama_arg(
|
||||
{"--no-mmap"},
|
||||
"do not memory-map model (slower load but may reduce pageouts if not using mlock)",
|
||||
[](gpt_params & params) {
|
||||
params.use_mmap = false;
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_NO_MMAP"));
|
||||
add_opt(llama_arg(
|
||||
{"--numa"}, "TYPE",
|
||||
"attempt optimizations that help on some NUMA systems\n"
|
||||
@ -1385,7 +1385,7 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
|
||||
else if (value == "numactl") { params.numa = GGML_NUMA_STRATEGY_NUMACTL; }
|
||||
else { throw std::invalid_argument("invalid value"); }
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_NUMA"));
|
||||
add_opt(llama_arg(
|
||||
{"-ngl", "--gpu-layers", "--n-gpu-layers"}, "N",
|
||||
"number of layers to store in VRAM",
|
||||
@ -1433,7 +1433,7 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
|
||||
fprintf(stderr, "warning: llama.cpp was compiled without support for GPU offload. Setting the split mode has no effect.\n");
|
||||
}
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_SPLIT_MODE"));
|
||||
add_opt(llama_arg(
|
||||
{"-ts", "--tensor-split"}, "N0,N1,N2,...",
|
||||
"fraction of the model to offload to each GPU, comma-separated list of proportions, e.g. 3,1",
|
||||
@ -1460,7 +1460,7 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
|
||||
fprintf(stderr, "warning: llama.cpp was compiled without support for GPU offload. Setting a tensor split has no effect.\n");
|
||||
}
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_TENSOR_SPLIT"));
|
||||
add_opt(llama_arg(
|
||||
{"-mg", "--main-gpu"}, "INDEX",
|
||||
format("the GPU to use for the model (with split-mode = none), or for intermediate results and KV (with split-mode = row) (default: %d)", params.main_gpu),
|
||||
@ -1470,7 +1470,7 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
|
||||
fprintf(stderr, "warning: llama.cpp was compiled without support for GPU offload. Setting the main GPU has no effect.\n");
|
||||
}
|
||||
}
|
||||
));
|
||||
).set_env("LLAMA_ARG_MAIN_GPU"));
|
||||
add_opt(llama_arg(
|
||||
{"--check-tensors"},
|
||||
format("check model tensor data for invalid values (default: %s)", params.check_tensors ? "true" : "false"),
|
||||
@ -1533,7 +1533,7 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
|
||||
[](gpt_params & params, const std::string & value) {
|
||||
params.model_alias = value;
|
||||
}
|
||||
).set_examples({LLAMA_EXAMPLE_SERVER}));
|
||||
).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_ALIAS"));
|
||||
add_opt(llama_arg(
|
||||
{"-m", "--model"}, "FNAME",
|
||||
ex == LLAMA_EXAMPLE_EXPORT_LORA
|
||||
@ -1741,7 +1741,7 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
|
||||
[](gpt_params & params, const std::string & value) {
|
||||
params.public_path = value;
|
||||
}
|
||||
).set_examples({LLAMA_EXAMPLE_SERVER}));
|
||||
).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_STATIC_PATH"));
|
||||
add_opt(llama_arg(
|
||||
{"--embedding", "--embeddings"},
|
||||
format("restrict to only support embedding use case; use only with dedicated embedding models (default: %s)", params.embedding ? "enabled" : "disabled"),
|
||||
@ -1779,14 +1779,14 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
|
||||
[](gpt_params & params, const std::string & value) {
|
||||
params.ssl_file_key = value;
|
||||
}
|
||||
).set_examples({LLAMA_EXAMPLE_SERVER}));
|
||||
).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_SSL_KEY_FILE"));
|
||||
add_opt(llama_arg(
|
||||
{"--ssl-cert-file"}, "FNAME",
|
||||
"path to file a PEM-encoded SSL certificate",
|
||||
[](gpt_params & params, const std::string & value) {
|
||||
params.ssl_file_cert = value;
|
||||
}
|
||||
).set_examples({LLAMA_EXAMPLE_SERVER}));
|
||||
).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_SSL_CERT_FILE"));
|
||||
add_opt(llama_arg(
|
||||
{"-to", "--timeout"}, "N",
|
||||
format("server read/write timeout in seconds (default: %d)", params.timeout_read),
|
||||
@ -1794,7 +1794,7 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
|
||||
params.timeout_read = value;
|
||||
params.timeout_write = value;
|
||||
}
|
||||
).set_examples({LLAMA_EXAMPLE_SERVER}));
|
||||
).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_TIMEOUT"));
|
||||
add_opt(llama_arg(
|
||||
{"--threads-http"}, "N",
|
||||
format("number of threads used to process HTTP requests (default: %d)", params.n_threads_http),
|
||||
|
@ -6,15 +6,12 @@
|
||||
|
||||
// Export usage message (-h) to markdown format
|
||||
|
||||
static void export_md(std::string fname, llama_example ex) {
|
||||
std::ofstream file(fname, std::ofstream::out | std::ofstream::trunc);
|
||||
|
||||
gpt_params params;
|
||||
auto ctx_arg = gpt_params_parser_init(params, ex);
|
||||
|
||||
static void write_table_header(std::ofstream & file) {
|
||||
file << "| Argument | Explanation |\n";
|
||||
file << "| -------- | ----------- |\n";
|
||||
for (auto & opt : ctx_arg.options) {
|
||||
}
|
||||
|
||||
static void write_table_entry(std::ofstream & file, const llama_arg & opt) {
|
||||
file << "| `";
|
||||
// args
|
||||
for (const auto & arg : opt.args) {
|
||||
@ -42,6 +39,40 @@ static void export_md(std::string fname, llama_example ex) {
|
||||
string_replace_all(md_help, "|", "\\|");
|
||||
file << "` | " << md_help << " |\n";
|
||||
}
|
||||
|
||||
static void write_table(std::ofstream & file, std::vector<llama_arg *> & opts) {
|
||||
write_table_header(file);
|
||||
for (const auto & opt : opts) {
|
||||
write_table_entry(file, *opt);
|
||||
}
|
||||
}
|
||||
|
||||
static void export_md(std::string fname, llama_example ex) {
|
||||
std::ofstream file(fname, std::ofstream::out | std::ofstream::trunc);
|
||||
|
||||
gpt_params params;
|
||||
auto ctx_arg = gpt_params_parser_init(params, ex);
|
||||
|
||||
std::vector<llama_arg *> common_options;
|
||||
std::vector<llama_arg *> sparam_options;
|
||||
std::vector<llama_arg *> specific_options;
|
||||
for (auto & opt : ctx_arg.options) {
|
||||
// in case multiple LLAMA_EXAMPLE_* are set, we prioritize the LLAMA_EXAMPLE_* matching current example
|
||||
if (opt.is_sparam) {
|
||||
sparam_options.push_back(&opt);
|
||||
} else if (opt.in_example(ctx_arg.ex)) {
|
||||
specific_options.push_back(&opt);
|
||||
} else {
|
||||
common_options.push_back(&opt);
|
||||
}
|
||||
}
|
||||
|
||||
file << "**Common params**\n\n";
|
||||
write_table(file, common_options);
|
||||
file << "\n\n**Sampling params**\n\n";
|
||||
write_table(file, sparam_options);
|
||||
file << "\n\n**Example-specific params**\n\n";
|
||||
write_table(file, specific_options);
|
||||
}
|
||||
|
||||
int main(int, char **) {
|
||||
|
@ -17,6 +17,8 @@ The project is under active development, and we are [looking for feedback and co
|
||||
|
||||
## Usage
|
||||
|
||||
**Common params**
|
||||
|
||||
| Argument | Explanation |
|
||||
| -------- | ----------- |
|
||||
| `-h, --help, --usage` | print usage and exit |
|
||||
@ -38,7 +40,6 @@ The project is under active development, and we are [looking for feedback and co
|
||||
| `-b, --batch-size N` | logical maximum batch size (default: 2048)<br/>(env: LLAMA_ARG_BATCH) |
|
||||
| `-ub, --ubatch-size N` | physical maximum batch size (default: 512)<br/>(env: LLAMA_ARG_UBATCH) |
|
||||
| `--keep N` | number of tokens to keep from the initial prompt (default: 0, -1 = all) |
|
||||
| `--no-context-shift` | disables context shift on inifinite text generation (default: disabled) |
|
||||
| `-fa, --flash-attn` | enable Flash Attention (default: disabled)<br/>(env: LLAMA_ARG_FLASH_ATTN) |
|
||||
| `-p, --prompt PROMPT` | prompt to start generation with |
|
||||
| `--no-perf` | disable internal libllama performance timings (default: false)<br/>(env: LLAMA_ARG_NO_PERF) |
|
||||
@ -46,8 +47,56 @@ The project is under active development, and we are [looking for feedback and co
|
||||
| `-bf, --binary-file FNAME` | binary file containing the prompt (default: none) |
|
||||
| `-e, --escape` | process escapes sequences (\n, \r, \t, \', \", \\) (default: true) |
|
||||
| `--no-escape` | do not process escape sequences |
|
||||
| `-sp, --special` | special tokens output enabled (default: false) |
|
||||
| `--spm-infill` | use Suffix/Prefix/Middle pattern for infill (instead of Prefix/Suffix/Middle) as some models prefer this. (default: disabled) |
|
||||
| `--rope-scaling {none,linear,yarn}` | RoPE frequency scaling method, defaults to linear unless specified by the model<br/>(env: LLAMA_ARG_ROPE_SCALING_TYPE) |
|
||||
| `--rope-scale N` | RoPE context scaling factor, expands context by a factor of N<br/>(env: LLAMA_ARG_ROPE_SCALE) |
|
||||
| `--rope-freq-base N` | RoPE base frequency, used by NTK-aware scaling (default: loaded from model)<br/>(env: LLAMA_ARG_ROPE_FREQ_BASE) |
|
||||
| `--rope-freq-scale N` | RoPE frequency scaling factor, expands context by a factor of 1/N<br/>(env: LLAMA_ARG_ROPE_FREQ_SCALE) |
|
||||
| `--yarn-orig-ctx N` | YaRN: original context size of model (default: 0 = model training context size)<br/>(env: LLAMA_ARG_YARN_ORIG_CTX) |
|
||||
| `--yarn-ext-factor N` | YaRN: extrapolation mix factor (default: -1.0, 0.0 = full interpolation)<br/>(env: LLAMA_ARG_YARN_EXT_FACTOR) |
|
||||
| `--yarn-attn-factor N` | YaRN: scale sqrt(t) or attention magnitude (default: 1.0)<br/>(env: LLAMA_ARG_YARN_ATTN_FACTOR) |
|
||||
| `--yarn-beta-slow N` | YaRN: high correction dim or alpha (default: 1.0)<br/>(env: LLAMA_ARG_YARN_BETA_SLOW) |
|
||||
| `--yarn-beta-fast N` | YaRN: low correction dim or beta (default: 32.0)<br/>(env: LLAMA_ARG_YARN_BETA_FAST) |
|
||||
| `-gan, --grp-attn-n N` | group-attention factor (default: 1)<br/>(env: LLAMA_ARG_GRP_ATTN_N) |
|
||||
| `-gaw, --grp-attn-w N` | group-attention width (default: 512.0)<br/>(env: LLAMA_ARG_GRP_ATTN_W) |
|
||||
| `-dkvc, --dump-kv-cache` | verbose print of the KV cache |
|
||||
| `-nkvo, --no-kv-offload` | disable KV offload<br/>(env: LLAMA_ARG_NO_KV_OFFLOAD) |
|
||||
| `-ctk, --cache-type-k TYPE` | KV cache data type for K (default: f16)<br/>(env: LLAMA_ARG_CACHE_TYPE_K) |
|
||||
| `-ctv, --cache-type-v TYPE` | KV cache data type for V (default: f16)<br/>(env: LLAMA_ARG_CACHE_TYPE_V) |
|
||||
| `-dt, --defrag-thold N` | KV cache defragmentation threshold (default: -1.0, < 0 - disabled)<br/>(env: LLAMA_ARG_DEFRAG_THOLD) |
|
||||
| `-np, --parallel N` | number of parallel sequences to decode (default: 1)<br/>(env: LLAMA_ARG_N_PARALLEL) |
|
||||
| `--mlock` | force system to keep model in RAM rather than swapping or compressing<br/>(env: LLAMA_ARG_MLOCK) |
|
||||
| `--no-mmap` | do not memory-map model (slower load but may reduce pageouts if not using mlock)<br/>(env: LLAMA_ARG_NO_MMAP) |
|
||||
| `--numa TYPE` | attempt optimizations that help on some NUMA systems<br/>- distribute: spread execution evenly over all nodes<br/>- isolate: only spawn threads on CPUs on the node that execution started on<br/>- numactl: use the CPU map provided by numactl<br/>if run without this previously, it is recommended to drop the system page cache before using this<br/>see https://github.com/ggerganov/llama.cpp/issues/1437<br/>(env: LLAMA_ARG_NUMA) |
|
||||
| `-ngl, --gpu-layers, --n-gpu-layers N` | number of layers to store in VRAM<br/>(env: LLAMA_ARG_N_GPU_LAYERS) |
|
||||
| `-sm, --split-mode {none,layer,row}` | how to split the model across multiple GPUs, one of:<br/>- none: use one GPU only<br/>- layer (default): split layers and KV across GPUs<br/>- row: split rows across GPUs<br/>(env: LLAMA_ARG_SPLIT_MODE) |
|
||||
| `-ts, --tensor-split N0,N1,N2,...` | fraction of the model to offload to each GPU, comma-separated list of proportions, e.g. 3,1<br/>(env: LLAMA_ARG_TENSOR_SPLIT) |
|
||||
| `-mg, --main-gpu INDEX` | the GPU to use for the model (with split-mode = none), or for intermediate results and KV (with split-mode = row) (default: 0)<br/>(env: LLAMA_ARG_MAIN_GPU) |
|
||||
| `--check-tensors` | check model tensor data for invalid values (default: false) |
|
||||
| `--override-kv KEY=TYPE:VALUE` | advanced option to override model metadata by key. may be specified multiple times.<br/>types: int, float, bool, str. example: --override-kv tokenizer.ggml.add_bos_token=bool:false |
|
||||
| `--lora FNAME` | path to LoRA adapter (can be repeated to use multiple adapters) |
|
||||
| `--lora-scaled FNAME SCALE` | path to LoRA adapter with user defined scaling (can be repeated to use multiple adapters) |
|
||||
| `--control-vector FNAME` | add a control vector<br/>note: this argument can be repeated to add multiple control vectors |
|
||||
| `--control-vector-scaled FNAME SCALE` | add a control vector with user defined scaling SCALE<br/>note: this argument can be repeated to add multiple scaled control vectors |
|
||||
| `--control-vector-layer-range START END` | layer range to apply the control vector(s) to, start and end inclusive |
|
||||
| `-m, --model FNAME` | model path (default: `models/$filename` with filename from `--hf-file` or `--model-url` if set, otherwise models/7B/ggml-model-f16.gguf)<br/>(env: LLAMA_ARG_MODEL) |
|
||||
| `-mu, --model-url MODEL_URL` | model download url (default: unused)<br/>(env: LLAMA_ARG_MODEL_URL) |
|
||||
| `-hfr, --hf-repo REPO` | Hugging Face model repository (default: unused)<br/>(env: LLAMA_ARG_HF_REPO) |
|
||||
| `-hff, --hf-file FILE` | Hugging Face model file (default: unused)<br/>(env: LLAMA_ARG_HF_FILE) |
|
||||
| `-hft, --hf-token TOKEN` | Hugging Face access token (default: value from HF_TOKEN environment variable)<br/>(env: HF_TOKEN) |
|
||||
| `-ld, --logdir LOGDIR` | path under which to save YAML logs (no logging if unset) |
|
||||
| `--log-disable` | Log disable |
|
||||
| `--log-file FNAME` | Log to file |
|
||||
| `--log-colors` | Enable colored logging<br/>(env: LLAMA_LOG_COLORS) |
|
||||
| `-v, --verbose, --log-verbose` | Set verbosity level to infinity (i.e. log all messages, useful for debugging) |
|
||||
| `-lv, --verbosity, --log-verbosity N` | Set the verbosity threshold. Messages with a higher verbosity will be ignored.<br/>(env: LLAMA_LOG_VERBOSITY) |
|
||||
| `--log-prefix` | Enable prefx in log messages<br/>(env: LLAMA_LOG_PREFIX) |
|
||||
| `--log-timestamps` | Enable timestamps in log messages<br/>(env: LLAMA_LOG_TIMESTAMPS) |
|
||||
|
||||
|
||||
**Sampling params**
|
||||
|
||||
| Argument | Explanation |
|
||||
| -------- | ----------- |
|
||||
| `--samplers SAMPLERS` | samplers that will be used for generation in the order, separated by ';'<br/>(default: top_k;tfs_z;typ_p;top_p;min_p;temperature) |
|
||||
| `-s, --seed SEED` | RNG seed (default: 4294967295, use random seed for 4294967295) |
|
||||
| `--sampling-seq SEQUENCE` | simplified sequence for samplers that will be used (default: kfypmt) |
|
||||
@ -72,54 +121,28 @@ The project is under active development, and we are [looking for feedback and co
|
||||
| `--grammar GRAMMAR` | BNF-like grammar to constrain generations (see samples in grammars/ dir) (default: '') |
|
||||
| `--grammar-file FNAME` | file to read grammar from |
|
||||
| `-j, --json-schema SCHEMA` | JSON schema to constrain generations (https://json-schema.org/), e.g. `{}` for any JSON object<br/>For schemas w/ external $refs, use --grammar + example/json_schema_to_grammar.py instead |
|
||||
| `--rope-scaling {none,linear,yarn}` | RoPE frequency scaling method, defaults to linear unless specified by the model |
|
||||
| `--rope-scale N` | RoPE context scaling factor, expands context by a factor of N |
|
||||
| `--rope-freq-base N` | RoPE base frequency, used by NTK-aware scaling (default: loaded from model) |
|
||||
| `--rope-freq-scale N` | RoPE frequency scaling factor, expands context by a factor of 1/N |
|
||||
| `--yarn-orig-ctx N` | YaRN: original context size of model (default: 0 = model training context size) |
|
||||
| `--yarn-ext-factor N` | YaRN: extrapolation mix factor (default: -1.0, 0.0 = full interpolation) |
|
||||
| `--yarn-attn-factor N` | YaRN: scale sqrt(t) or attention magnitude (default: 1.0) |
|
||||
| `--yarn-beta-slow N` | YaRN: high correction dim or alpha (default: 1.0) |
|
||||
| `--yarn-beta-fast N` | YaRN: low correction dim or beta (default: 32.0) |
|
||||
| `-gan, --grp-attn-n N` | group-attention factor (default: 1) |
|
||||
| `-gaw, --grp-attn-w N` | group-attention width (default: 512.0) |
|
||||
| `-dkvc, --dump-kv-cache` | verbose print of the KV cache |
|
||||
| `-nkvo, --no-kv-offload` | disable KV offload |
|
||||
| `-ctk, --cache-type-k TYPE` | KV cache data type for K (default: f16) |
|
||||
| `-ctv, --cache-type-v TYPE` | KV cache data type for V (default: f16) |
|
||||
| `-dt, --defrag-thold N` | KV cache defragmentation threshold (default: -1.0, < 0 - disabled)<br/>(env: LLAMA_ARG_DEFRAG_THOLD) |
|
||||
| `-np, --parallel N` | number of parallel sequences to decode (default: 1)<br/>(env: LLAMA_ARG_N_PARALLEL) |
|
||||
|
||||
|
||||
**Example-specific params**
|
||||
|
||||
| Argument | Explanation |
|
||||
| -------- | ----------- |
|
||||
| `--no-context-shift` | disables context shift on inifinite text generation (default: disabled)<br/>(env: LLAMA_ARG_NO_CONTEXT_SHIFT) |
|
||||
| `-sp, --special` | special tokens output enabled (default: false) |
|
||||
| `--spm-infill` | use Suffix/Prefix/Middle pattern for infill (instead of Prefix/Suffix/Middle) as some models prefer this. (default: disabled) |
|
||||
| `--pooling {none,mean,cls,last}` | pooling type for embeddings, use model default if unspecified<br/>(env: LLAMA_ARG_POOLING) |
|
||||
| `-cb, --cont-batching` | enable continuous batching (a.k.a dynamic batching) (default: enabled)<br/>(env: LLAMA_ARG_CONT_BATCHING) |
|
||||
| `-nocb, --no-cont-batching` | disable continuous batching<br/>(env: LLAMA_ARG_NO_CONT_BATCHING) |
|
||||
| `--mlock` | force system to keep model in RAM rather than swapping or compressing |
|
||||
| `--no-mmap` | do not memory-map model (slower load but may reduce pageouts if not using mlock) |
|
||||
| `--numa TYPE` | attempt optimizations that help on some NUMA systems<br/>- distribute: spread execution evenly over all nodes<br/>- isolate: only spawn threads on CPUs on the node that execution started on<br/>- numactl: use the CPU map provided by numactl<br/>if run without this previously, it is recommended to drop the system page cache before using this<br/>see https://github.com/ggerganov/llama.cpp/issues/1437 |
|
||||
| `-ngl, --gpu-layers, --n-gpu-layers N` | number of layers to store in VRAM<br/>(env: LLAMA_ARG_N_GPU_LAYERS) |
|
||||
| `-sm, --split-mode {none,layer,row}` | how to split the model across multiple GPUs, one of:<br/>- none: use one GPU only<br/>- layer (default): split layers and KV across GPUs<br/>- row: split rows across GPUs |
|
||||
| `-ts, --tensor-split N0,N1,N2,...` | fraction of the model to offload to each GPU, comma-separated list of proportions, e.g. 3,1 |
|
||||
| `-mg, --main-gpu INDEX` | the GPU to use for the model (with split-mode = none), or for intermediate results and KV (with split-mode = row) (default: 0) |
|
||||
| `--check-tensors` | check model tensor data for invalid values (default: false) |
|
||||
| `--override-kv KEY=TYPE:VALUE` | advanced option to override model metadata by key. may be specified multiple times.<br/>types: int, float, bool, str. example: --override-kv tokenizer.ggml.add_bos_token=bool:false |
|
||||
| `--lora FNAME` | path to LoRA adapter (can be repeated to use multiple adapters) |
|
||||
| `--lora-scaled FNAME SCALE` | path to LoRA adapter with user defined scaling (can be repeated to use multiple adapters) |
|
||||
| `--control-vector FNAME` | add a control vector<br/>note: this argument can be repeated to add multiple control vectors |
|
||||
| `--control-vector-scaled FNAME SCALE` | add a control vector with user defined scaling SCALE<br/>note: this argument can be repeated to add multiple scaled control vectors |
|
||||
| `--control-vector-layer-range START END` | layer range to apply the control vector(s) to, start and end inclusive |
|
||||
| `-a, --alias STRING` | set alias for model name (to be used by REST API) |
|
||||
| `-m, --model FNAME` | model path (default: `models/$filename` with filename from `--hf-file` or `--model-url` if set, otherwise models/7B/ggml-model-f16.gguf)<br/>(env: LLAMA_ARG_MODEL) |
|
||||
| `-mu, --model-url MODEL_URL` | model download url (default: unused)<br/>(env: LLAMA_ARG_MODEL_URL) |
|
||||
| `-hfr, --hf-repo REPO` | Hugging Face model repository (default: unused)<br/>(env: LLAMA_ARG_HF_REPO) |
|
||||
| `-hff, --hf-file FILE` | Hugging Face model file (default: unused)<br/>(env: LLAMA_ARG_HF_FILE) |
|
||||
| `-hft, --hf-token TOKEN` | Hugging Face access token (default: value from HF_TOKEN environment variable)<br/>(env: HF_TOKEN) |
|
||||
| `-a, --alias STRING` | set alias for model name (to be used by REST API)<br/>(env: LLAMA_ARG_ALIAS) |
|
||||
| `--host HOST` | ip address to listen (default: 127.0.0.1)<br/>(env: LLAMA_ARG_HOST) |
|
||||
| `--port PORT` | port to listen (default: 8080)<br/>(env: LLAMA_ARG_PORT) |
|
||||
| `--path PATH` | path to serve static files from (default: ) |
|
||||
| `--path PATH` | path to serve static files from (default: )<br/>(env: LLAMA_ARG_STATIC_PATH) |
|
||||
| `--embedding, --embeddings` | restrict to only support embedding use case; use only with dedicated embedding models (default: disabled)<br/>(env: LLAMA_ARG_EMBEDDINGS) |
|
||||
| `--api-key KEY` | API key to use for authentication (default: none)<br/>(env: LLAMA_API_KEY) |
|
||||
| `--api-key-file FNAME` | path to file containing API keys (default: none) |
|
||||
| `--ssl-key-file FNAME` | path to file a PEM-encoded SSL private key |
|
||||
| `--ssl-cert-file FNAME` | path to file a PEM-encoded SSL certificate |
|
||||
| `-to, --timeout N` | server read/write timeout in seconds (default: 600) |
|
||||
| `--ssl-key-file FNAME` | path to file a PEM-encoded SSL private key<br/>(env: LLAMA_ARG_SSL_KEY_FILE) |
|
||||
| `--ssl-cert-file FNAME` | path to file a PEM-encoded SSL certificate<br/>(env: LLAMA_ARG_SSL_CERT_FILE) |
|
||||
| `-to, --timeout N` | server read/write timeout in seconds (default: 600)<br/>(env: LLAMA_ARG_TIMEOUT) |
|
||||
| `--threads-http N` | number of threads used to process HTTP requests (default: -1)<br/>(env: LLAMA_ARG_THREADS_HTTP) |
|
||||
| `-spf, --system-prompt-file FNAME` | set a file to load a system prompt (initial prompt of all slots), this is useful for chat applications |
|
||||
| `--metrics` | enable prometheus compatible metrics endpoint (default: disabled)<br/>(env: LLAMA_ARG_ENDPOINT_METRICS) |
|
||||
@ -128,14 +151,6 @@ The project is under active development, and we are [looking for feedback and co
|
||||
| `--chat-template JINJA_TEMPLATE` | set custom jinja chat template (default: template taken from model's metadata)<br/>if suffix/prefix are specified, template will be disabled<br/>only commonly used templates are accepted:<br/>https://github.com/ggerganov/llama.cpp/wiki/Templates-supported-by-llama_chat_apply_template<br/>(env: LLAMA_ARG_CHAT_TEMPLATE) |
|
||||
| `-sps, --slot-prompt-similarity SIMILARITY` | how much the prompt of a request must match the prompt of a slot in order to use that slot (default: 0.50, 0.0 = disabled)<br/> |
|
||||
| `--lora-init-without-apply` | load LoRA adapters without applying them (apply later via POST /lora-adapters) (default: disabled) |
|
||||
| `-ld, --logdir LOGDIR` | path under which to save YAML logs (no logging if unset) |
|
||||
| `--log-disable` | Log disable |
|
||||
| `--log-file FNAME` | Log to file |
|
||||
| `--log-colors` | Enable colored logging<br/>(env: LLAMA_LOG_COLORS) |
|
||||
| `-v, --verbose, --log-verbose` | Set verbosity level to infinity (i.e. log all messages, useful for debugging) |
|
||||
| `-lv, --verbosity, --log-verbosity N` | Set the verbosity threshold. Messages with a higher verbosity will be ignored.<br/>(env: LLAMA_LOG_VERBOSITY) |
|
||||
| `--log-prefix` | Enable prefx in log messages<br/>(env: LLAMA_LOG_PREFIX) |
|
||||
| `--log-timestamps` | Enable timestamps in log messages<br/>(env: LLAMA_LOG_TIMESTAMPS) |
|
||||
|
||||
Note: If both command line argument and environment variable are both set for the same param, the argument will take precedence over env var.
|
||||
|
||||
|
@ -2356,6 +2356,10 @@ int main(int argc, char ** argv) {
|
||||
svr.reset(new httplib::Server());
|
||||
}
|
||||
#else
|
||||
if (params.ssl_file_key != "" && params.ssl_file_cert != "") {
|
||||
LOG_ERR("Server is built without SSL support\n");
|
||||
return 1;
|
||||
}
|
||||
svr.reset(new httplib::Server());
|
||||
#endif
|
||||
|
||||
|
Loading…
Reference in New Issue
Block a user