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

token healing : fix rebase bug

Browse source
This commit is contained in:
mare5x 2024-07-08 16:18:19 +02:00
parent 940ab81784
commit b27f87d6da
1 changed files with 3 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

6
common/sampling.cpp
View file

@ -568,8 +568,6 @@ static llama_token_data_array llama_sampling_prepare_impl(
llama_sample_apply_guidance(ctx_main, logits, logits_guidance, params.cfg_scale); llama_sample_apply_guidance(ctx_main, logits, logits_guidance, params.cfg_scale);
} }
cur.resize(n_vocab);
// Constrain tokens based on the remaining token healing prefix (if any) // Constrain tokens based on the remaining token healing prefix (if any)
const auto & th_prefix = ctx_sampling->token_healing_prefix; const auto & th_prefix = ctx_sampling->token_healing_prefix;
if (params.token_healing.enabled && !th_prefix.empty()) { if (params.token_healing.enabled && !th_prefix.empty()) {
@ -583,10 +581,12 @@ static llama_token_data_array llama_sampling_prepare_impl(
} }
// N.B. We could also set token constraints by setting rejected tokens' logits to -inf // N.B. We could also set token constraints by setting rejected tokens' logits to -inf
cur.clear();
for (const llama_token token_id : th_candidates) { for (const llama_token token_id : th_candidates) {
cur[token_id] = llama_token_data{token_id, logits[token_id], 0.0f}; cur.emplace_back(llama_token_data{token_id, logits[token_id], 0.0f});
} }
} else { } else {
cur.resize(n_vocab);
for (llama_token token_id = 0; token_id < n_vocab; token_id++) { for (llama_token token_id = 0; token_id < n_vocab; token_id++) {
cur[token_id] = llama_token_data{token_id, logits[token_id], 0.0f}; cur[token_id] = llama_token_data{token_id, logits[token_id], 0.0f};
} }