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

llama : avoid hardcoded special tokens

Browse source
This commit is contained in:
Georgi Gerganov 2023-08-18 17:29:20 +03:00
parent 035d511457
commit 5d2656d670
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
11 changed files with 61 additions and 65 deletions
Download patch file Download diff file Expand all files Collapse all files

8
common/common.cpp
View file

@ -427,7 +427,7 @@ bool gpt_params_parse(int argc, char ** argv, gpt_params & params) {
}
params.hellaswag_tasks = std::stoi(argv[i]);
} else if (arg == "--ignore-eos") {
params.logit_bias[llama_token_eos()] = -INFINITY;
params.ignore_eos = true;
} else if (arg == "--no-penalize-nl") {
params.penalize_nl = false;
} else if (arg == "-l" || arg == "--logit-bias") {
@ -662,7 +662,7 @@ struct llama_context_params llama_context_params_from_gpt_params(const gpt_param
return lparams;
}
std::tuple<struct llama_model *, struct llama_context *> llama_init_from_gpt_params(const gpt_params & params) {
std::tuple<struct llama_model *, struct llama_context *> llama_init_from_gpt_params(gpt_params & params) {
auto lparams = llama_context_params_from_gpt_params(params);
llama_model * model = llama_load_model_from_file(params.model.c_str(), lparams);
@ -691,6 +691,10 @@ std::tuple<struct llama_model *, struct llama_context *> llama_init_from_gpt_par
}
}
if (params.ignore_eos) {
params.logit_bias[llama_token_eos(lctx)] = -INFINITY;
}
return std::make_tuple(model, lctx);
}