diff --git a/CODEOWNERS b/CODEOWNERS deleted file mode 100644 index 88ab6de4f..000000000 --- a/CODEOWNERS +++ /dev/null @@ -1,3 +0,0 @@ -# collaborators can optionally add themselves here to indicate their availability for reviewing related PRs - -ci/ @ggerganov diff --git a/common/arg.cpp b/common/arg.cpp index 4fcf6167a..cc0bb5621 100644 --- a/common/arg.cpp +++ b/common/arg.cpp @@ -146,6 +146,35 @@ static void common_params_handle_model_default(common_params & params) { } } +const std::vector kv_cache_types = { + GGML_TYPE_F32, + GGML_TYPE_F16, + GGML_TYPE_BF16, + GGML_TYPE_Q8_0, + GGML_TYPE_Q4_0, + GGML_TYPE_Q4_1, + GGML_TYPE_IQ4_NL, + GGML_TYPE_Q5_0, + GGML_TYPE_Q5_1, +}; + +static ggml_type kv_cache_type_from_str(const std::string & s) { + for (const auto & type : kv_cache_types) { + if (ggml_type_name(type) == s) { + return type; + } + } + throw std::runtime_error("Unsupported cache type: " + s); +} + +static std::string get_all_kv_cache_types() { + std::ostringstream msg; + for (const auto & type : kv_cache_types) { + msg << ggml_type_name(type) << (&type == &kv_cache_types.back() ? "" : ", "); + } + return msg.str(); +} + // // CLI argument parsing functions // @@ -1175,18 +1204,28 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_env("LLAMA_ARG_NO_KV_OFFLOAD")); add_opt(common_arg( {"-ctk", "--cache-type-k"}, "TYPE", - string_format("KV cache data type for K (default: %s)", params.cache_type_k.c_str()), + string_format( + "KV cache data type for K\n" + "allowed values: %s\n" + "(default: %s)", + get_all_kv_cache_types().c_str(), + ggml_type_name(params.cache_type_k) + ), [](common_params & params, const std::string & value) { - // TODO: get the type right here - params.cache_type_k = value; + params.cache_type_k = kv_cache_type_from_str(value); } ).set_env("LLAMA_ARG_CACHE_TYPE_K")); add_opt(common_arg( {"-ctv", "--cache-type-v"}, "TYPE", - string_format("KV cache data type for V (default: %s)", params.cache_type_v.c_str()), + string_format( + "KV cache data type for V\n" + "allowed values: %s\n" + "(default: %s)", + get_all_kv_cache_types().c_str(), + ggml_type_name(params.cache_type_v) + ), [](common_params & params, const std::string & value) { - // TODO: get the type right here - params.cache_type_v = value; + params.cache_type_v = kv_cache_type_from_str(value); } ).set_env("LLAMA_ARG_CACHE_TYPE_V")); add_opt(common_arg( @@ -2084,35 +2123,35 @@ common_params_context common_params_parser_init(common_params & params, llama_ex [](common_params & params, int value) { params.speculative.n_max = value; } - ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_LOOKUP, LLAMA_EXAMPLE_SERVER})); + ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_LOOKUP, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_DRAFT_MAX")); add_opt(common_arg( {"--draft-min", "--draft-n-min"}, "N", string_format("minimum number of draft tokens to use for speculative decoding (default: %d)", params.speculative.n_min), [](common_params & params, int value) { params.speculative.n_min = value; } - ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_LOOKUP, LLAMA_EXAMPLE_SERVER})); + ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_LOOKUP, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_DRAFT_MIN")); add_opt(common_arg( {"--draft-p-split"}, "P", string_format("speculative decoding split probability (default: %.1f)", (double)params.speculative.p_split), [](common_params & params, const std::string & value) { params.speculative.p_split = std::stof(value); } - ).set_examples({LLAMA_EXAMPLE_SPECULATIVE})); + ).set_examples({LLAMA_EXAMPLE_SPECULATIVE}).set_env("LLAMA_ARG_DRAFT_P_SPLIT")); add_opt(common_arg( {"--draft-p-min"}, "P", string_format("minimum speculative decoding probability (greedy) (default: %.1f)", (double)params.speculative.p_min), [](common_params & params, const std::string & value) { params.speculative.p_min = std::stof(value); } - ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER})); + ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_DRAFT_P_MIN")); add_opt(common_arg( {"-cd", "--ctx-size-draft"}, "N", string_format("size of the prompt context for the draft model (default: %d, 0 = loaded from model)", params.speculative.n_ctx), [](common_params & params, int value) { params.speculative.n_ctx = value; } - ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER})); + ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_CTX_SIZE_DRAFT")); add_opt(common_arg( {"-devd", "--device-draft"}, "", "comma-separated list of devices to use for offloading the draft model (none = don't offload)\n" @@ -2132,14 +2171,14 @@ common_params_context common_params_parser_init(common_params & params, llama_ex fprintf(stderr, "warning: consult docs/build.md for compilation instructions\n"); } } - ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER})); + ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_N_GPU_LAYERS_DRAFT")); add_opt(common_arg( {"-md", "--model-draft"}, "FNAME", "draft model for speculative decoding (default: unused)", [](common_params & params, const std::string & value) { params.speculative.model = value; } - ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER})); + ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_MODEL_DRAFT")); return ctx_arg; } diff --git a/common/common.cpp b/common/common.cpp index b37332891..9ff1a2965 100644 --- a/common/common.cpp +++ b/common/common.cpp @@ -1017,38 +1017,6 @@ struct llama_model_params common_model_params_to_llama(common_params & params) { return mparams; } -static ggml_type kv_cache_type_from_str(const std::string & s) { - if (s == "f32") { - return GGML_TYPE_F32; - } - if (s == "f16") { - return GGML_TYPE_F16; - } - if (s == "bf16") { - return GGML_TYPE_BF16; - } - if (s == "q8_0") { - return GGML_TYPE_Q8_0; - } - if (s == "q4_0") { - return GGML_TYPE_Q4_0; - } - if (s == "q4_1") { - return GGML_TYPE_Q4_1; - } - if (s == "iq4_nl") { - return GGML_TYPE_IQ4_NL; - } - if (s == "q5_0") { - return GGML_TYPE_Q5_0; - } - if (s == "q5_1") { - return GGML_TYPE_Q5_1; - } - - throw std::runtime_error("Unsupported cache type: " + s); -} - struct llama_context_params common_context_params_to_llama(const common_params & params) { auto cparams = llama_context_default_params(); @@ -1083,8 +1051,8 @@ struct llama_context_params common_context_params_to_llama(const common_params & cparams.pooling_type = LLAMA_POOLING_TYPE_RANK; } - cparams.type_k = kv_cache_type_from_str(params.cache_type_k); - cparams.type_v = kv_cache_type_from_str(params.cache_type_v); + cparams.type_k = params.cache_type_k; + cparams.type_v = params.cache_type_v; return cparams; } diff --git a/common/common.h b/common/common.h index 3da9a6e9d..1181cd4c5 100644 --- a/common/common.h +++ b/common/common.h @@ -282,8 +282,8 @@ struct common_params { bool warmup = true; // warmup run bool check_tensors = false; // validate tensor data - std::string cache_type_k = "f16"; // KV cache data type for the K - std::string cache_type_v = "f16"; // KV cache data type for the V + ggml_type cache_type_k = GGML_TYPE_F16; // KV cache data type for the K + ggml_type cache_type_v = GGML_TYPE_F16; // KV cache data type for the V // multimodal models (see examples/llava) std::string mmproj = ""; // path to multimodal projector // NOLINT diff --git a/examples/gguf-split/gguf-split.cpp b/examples/gguf-split/gguf-split.cpp index 7a9ab0000..ca17c2782 100644 --- a/examples/gguf-split/gguf-split.cpp +++ b/examples/gguf-split/gguf-split.cpp @@ -288,7 +288,7 @@ struct split_strategy { } void print_info() { - printf("n_split: %ld\n", ctx_outs.size()); + printf("n_split: %zu\n", ctx_outs.size()); int i_split = 0; for (auto & ctx_out : ctx_outs) { // re-calculate the real gguf size for each split (= metadata size + total size of all tensors) @@ -298,7 +298,7 @@ struct split_strategy { total_size += ggml_nbytes(t); } total_size = total_size / 1000 / 1000; // convert to megabytes - printf("split %05d: n_tensors = %d, total_size = %ldM\n", i_split + 1, gguf_get_n_tensors(ctx_out), total_size); + printf("split %05d: n_tensors = %d, total_size = %zuM\n", i_split + 1, gguf_get_n_tensors(ctx_out), total_size); i_split++; } } diff --git a/examples/quantize/README.md b/examples/quantize/README.md index 5d1e11c67..f9cce7b21 100644 --- a/examples/quantize/README.md +++ b/examples/quantize/README.md @@ -81,7 +81,7 @@ Several quantization methods are supported. They differ in the resulting model d - [#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) + - [#4996 - k-quants 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) diff --git a/examples/server/public/index.html b/examples/server/public/index.html index 250729a44..9a19c5e83 100644 --- a/examples/server/public/index.html +++ b/examples/server/public/index.html @@ -11,84 +11,84 @@ 🦙 llama.cpp - chat - - + @@ -99,7 +99,7 @@ Server rendered element contains fewer child nodes than client vdom.`),T=!0),au(
-
+

Conversations

@@ -204,51 +204,25 @@ Server rendered element contains fewer child nodes than client vdom.`),T=!0),au( {{ messages.length === 0 ? 'Send a message to start' : '' }}
-
-
- - - - -
-
- - -
- - - - - -
+
-
-
- - -
+
+
@@ -311,6 +285,10 @@ Server rendered element contains fewer child nodes than client vdom.`),T=!0),au(
Advanced config
+
+ + Show tokens per second +
+ + + + +