Skip to content

speculative : update default params #11954

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Feb 19, 2025
Merged

speculative : update default params #11954

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
965ad1c
speculative : update default params
ggerganov Feb 19, 2025
0f2bf55
speculative : do not discard the last drafted token
ggerganov Feb 19, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions common/common.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -178,10 +178,10 @@ struct common_params_speculative {

int32_t n_ctx = 0; // draft context size
int32_t n_max = 16; // maximum number of tokens to draft during speculative decoding
int32_t n_min = 5; // minimum number of draft tokens to use for speculative decoding
int32_t n_min = 0; // minimum number of draft tokens to use for speculative decoding
int32_t n_gpu_layers = -1; // number of layers to store in VRAM for the draft model (-1 - use default)
float p_split = 0.1f; // speculative decoding split probability
float p_min = 0.9f; // minimum speculative decoding probability (greedy)
float p_min = 0.75f; // minimum speculative decoding probability (greedy)

struct cpu_params cpuparams;
struct cpu_params cpuparams_batch;
Expand Down
10 changes: 5 additions & 5 deletions common/speculative.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -252,11 +252,6 @@ llama_tokens common_speculative_gen_draft(
// add drafted token for each sequence
const llama_token id = cur_p->data[0].id;

// only collect very high-confidence draft tokens
if (cur_p->data[0].p < params.p_min) {
break;
}

common_sampler_accept(smpl, id, true);

result.push_back(id);
Expand All @@ -265,6 +260,11 @@ llama_tokens common_speculative_gen_draft(
break;
}

// only collect very high-confidence draft tokens
if (cur_p->data[0].p < params.p_min) {
break;
}

common_batch_add(batch, id, n_past + i + 1, { 0 }, true);

// evaluate the drafted tokens on the draft model
Expand Down
2 changes: 1 addition & 1 deletion common/speculative.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,7 +9,7 @@ struct common_speculative_params {
int n_draft = 16; // max drafted tokens
int n_reuse = 256;

float p_min = 0.9f; // min probability required to accept a token in the draft
float p_min = 0.75f; // min probability required to accept a token in the draft
};

struct common_speculative * common_speculative_init(struct llama_context * ctx_dft);
Expand Down
2 changes: 1 addition & 1 deletion examples/server/server.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -274,7 +274,7 @@ struct server_task {
params.speculative.p_min = json_value(data, "speculative.p_min", defaults.speculative.p_min);

params.speculative.n_min = std::min(params.speculative.n_max, params.speculative.n_min);
params.speculative.n_min = std::max(params.speculative.n_min, 2);
params.speculative.n_min = std::max(params.speculative.n_min, 0);
params.speculative.n_max = std::max(params.speculative.n_max, 0);

// Use OpenAI API logprobs only if n_probs wasn't provided
Expand Down
Loading