vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

sampling : refactor + optimize penalties sampler (#10803)

Browse source 
* sampling : refactor + optimize penalties sampler

ggml-ci

* common : apply ignore_eos as logit bias

ggml-ci

* batched : remove penalties sampler

* params : allow penalty_last_n == -1 to be equal to context size

ggml-ci

* common : by default, move the penalties at the end of the sampling chain

ggml-ci

* common : ignore all EOG tokens

Co-authored-by: Diego Devesa <slarengh@gmail.com>

* common : move back the penalties at the front of the sampling chain

ggml-ci

* readme : restore hint about --ignore-eos flag [no ci]

* llama : minor

ggml-ci

* webui : update

---------

Co-authored-by: Diego Devesa <slarengh@gmail.com>
This commit is contained in:
Georgi Gerganov 2024-12-16 12:31:14 +02:00 • committed by GitHub
parent 4ddd199f6f
commit 644fd71b44
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
17 changed files with 111 additions and 152 deletions
Download patch file Download diff file Expand all files Collapse all files

15
common/common.h
View file

@ -95,6 +95,7 @@ enum common_sampler_type {
COMMON_SAMPLER_TYPE_TEMPERATURE = 7,
COMMON_SAMPLER_TYPE_XTC = 8,
COMMON_SAMPLER_TYPE_INFILL = 9,
COMMON_SAMPLER_TYPE_PENALTIES = 10,
};
// dimensionality reduction methods, used by cvector-generator
@ -130,7 +131,6 @@ struct common_params_sampling {
int32_t mirostat = 0; // 0 = disabled, 1 = mirostat, 2 = mirostat 2.0
float mirostat_tau = 5.00f; // target entropy
float mirostat_eta = 0.10f; // learning rate
bool penalize_nl = false; // consider newlines as a repeatable token
bool ignore_eos = false;
bool no_perf = false; // disable performance metrics
bool timing_per_token = false;
@ -139,6 +139,7 @@ struct common_params_sampling {
std::vector<enum common_sampler_type> samplers = {
COMMON_SAMPLER_TYPE_PENALTIES,
COMMON_SAMPLER_TYPE_DRY,
COMMON_SAMPLER_TYPE_TOP_K,
COMMON_SAMPLER_TYPE_TYPICAL_P,
@ -193,11 +194,13 @@ struct common_params {
float defrag_thold = 0.1f; // KV cache defragmentation threshold
// offload params
std::vector<ggml_backend_dev_t> devices; // devices to use for offloading
int32_t n_gpu_layers = -1; // number of layers to store in VRAM (-1 - use default)
int32_t main_gpu = 0; // the GPU that is used for scratch and small tensors
float tensor_split[128] = {0}; // how split tensors should be distributed across GPUs
enum llama_split_mode split_mode = LLAMA_SPLIT_MODE_LAYER; // how to split the model across GPUs
std::vector<ggml_backend_dev_t> devices; // devices to use for offloading
int32_t n_gpu_layers = -1; // number of layers to store in VRAM (-1 - use default)
int32_t main_gpu = 0; // the GPU that is used for scratch and small tensors
float tensor_split[128] = {0}; // how split tensors should be distributed across GPUs
enum llama_split_mode split_mode = LLAMA_SPLIT_MODE_LAYER; // how to split the model across GPUs
struct cpu_params cpuparams;
struct cpu_params cpuparams_batch;