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

common : fix mirostat state when using multiple sequences (#3543)

Browse source 
* Fix mirostat state when using multiple sequences

* Fix mirostat by completely refactoring sampling!

* Try to fix zig build.

* Export function to fetch/create default sampler states

Code formatting cleanups and add some comments

Silence a warning about id not being used when logging is disabled

* Apply some renaming suggestions.

Fix comments that were out of sync with the pull.

* Use more consistant naming convention for sampling contexts
This commit is contained in:
Kerfuffle 2023-10-11 13:35:46 -06:00 • committed by GitHub
parent 8c70a5ff25
commit 70c29da118
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
14 changed files with 495 additions and 334 deletions
Download patch file Download diff file Expand all files Collapse all files

6
examples/parallel/parallel.cpp
View file

@ -125,6 +125,8 @@ int main(int argc, char ** argv) {
params.logits_all = true;
std::tie(model, ctx) = llama_init_from_gpt_params(params);
llama_sampling_context ctx_sampling = llama_sampling_context_init(params, NULL);
// load the prompts from an external file if there are any
if (params.prompt.empty()) {
printf("\n\033[32mNo new questions so proceed with build-in defaults.\033[0m\n");
@ -339,7 +341,7 @@ int main(int argc, char ** argv) {
//printf("client %d, seq %d, token %d, pos %d, batch %d\n",
// client.id, client.seq_id, client.sampled, client.n_decoded, client.i_batch);
const llama_token id = llama_sample_token(ctx, NULL, NULL, params, client.tokens_prev, candidates, client.i_batch - i);
const llama_token id = llama_sampling_sample(ctx, NULL, ctx_sampling, client.tokens_prev, candidates, client.i_batch - i, client.seq_id);
if (client.n_decoded == 1) {
// start measuring generation time after the first token to make sure all concurrent clients
@ -384,7 +386,7 @@ int main(int argc, char ** argv) {
n_total_prompt += client.n_prompt;
n_total_gen += client.n_decoded;
llama_sampling_context_reset(ctx_sampling, client.seq_id);
client.seq_id = -1;
}