sampler : remove "full-context windows" from history-based samplers (#26524)

* Resolve -1 to 1024 instead of ctx-len for samplers

Because of backend-sampling we initialize samplers before the complete
llama_context is there. Therefore, we cannot infer the resolved context
length yet at the time we construct the samplers.

* Shared default of 64 for history-based samplers, remove context_size
This commit is contained in:
Oliver Simons
2026-08-04 21:28:55 +03:00
committed by GitHub
parent 76c956c137
commit a6aa6f5450
18 changed files with 52 additions and 80 deletions
+5 -5
View File
@@ -133,14 +133,14 @@ For the full list of features, please refer to [server's changelog](https://gith
| `--xtc-probability N` | xtc probability (default: 0.00, 0.0 = disabled) |
| `--xtc-threshold N` | xtc threshold (default: 0.10, 1.0 = disabled) |
| `--typical, --typical-p N` | locally typical sampling, parameter p (default: 1.00, 1.0 = disabled) |
| `--repeat-last-n N` | last n tokens to consider for penalize (default: 64, 0 = disabled, -1 = ctx_size) |
| `--repeat-last-n N` | last n tokens to consider for penalize (default: 64, 0 = disabled) |
| `--repeat-penalty N` | penalize repeat sequence of tokens (default: 1.00, 1.0 = disabled) |
| `--presence-penalty N` | repeat alpha presence penalty (default: 0.00, 0.0 = disabled) |
| `--frequency-penalty N` | repeat alpha frequency penalty (default: 0.00, 0.0 = disabled) |
| `--dry-multiplier N` | set DRY sampling multiplier (default: 0.00, 0.0 = disabled) |
| `--dry-base N` | set DRY sampling base value (default: 1.75) |
| `--dry-allowed-length N` | set allowed length for DRY sampling (default: 2) |
| `--dry-penalty-last-n N` | set DRY penalty for the last n tokens (default: -1, 0 = disable, -1 = context size) |
| `--dry-penalty-last-n N` | set DRY penalty for the last n tokens (default: 64, 0 = disable) |
| `--dry-sequence-breaker STRING` | add sequence breaker for DRY sampling, clearing out default breakers ('\n', ':', '"', '*') in the process; use "none" to not use any sequence breakers |
| `--adaptive-target N` | adaptive-p: select tokens near this probability (valid range 0.0 to 1.0; negative = disabled) (default: -1.00)<br/>[(more info)](https://github.com/ggml-org/llama.cpp/pull/17927) |
| `--adaptive-decay N` | adaptive-p: decay rate for target adaptation over time. lower values are more reactive, higher values are more stable.<br/>(valid range 0.0 to 0.99) (default: 0.90) |
@@ -476,7 +476,7 @@ These words will not be included in the completion, so make sure to add them to
`repeat_penalty`: Control the repetition of token sequences in the generated text. Default: `1.1`
`repeat_last_n`: Last n tokens to consider for penalizing repetition. Default: `64`, where `0` is disabled and `-1` is ctx-size.
`repeat_last_n`: Last n tokens to consider for penalizing repetition. Default: `64`, where `0` is disabled.
`presence_penalty`: Repeat alpha presence penalty. Default: `0.0`, which is disabled.
@@ -488,7 +488,7 @@ These words will not be included in the completion, so make sure to add them to
`dry_allowed_length`: Tokens that extend repetition beyond this receive exponentially increasing penalty: multiplier * base ^ (length of repeating sequence before token - allowed length). Default: `2`
`dry_penalty_last_n`: How many tokens to scan for repetitions. Default: `-1`, where `0` is disabled and `-1` is context size.
`dry_penalty_last_n`: How many tokens to scan for repetitions. Default: `64`, where `0` is disabled.
`dry_sequence_breakers`: Specify an array of sequence breakers for DRY sampling. Only a JSON array of strings is accepted. Default: `['\n', ':', '"', '*']`
@@ -796,7 +796,7 @@ By default, it is read-only. To make POST request to change global properties, y
"dry_multiplier": 0.0,
"dry_base": 1.75,
"dry_allowed_length": 2,
"dry_penalty_last_n": -1,
"dry_penalty_last_n": 64,
"dry_sequence_breakers": [
"\n",
":",
+1 -3
View File
@@ -1807,8 +1807,7 @@ private:
// initialize samplers
if (task.need_sampling()) {
try {
slot.smpl.reset(common_sampler_init(
model_tgt, task.params.sampling, (int32_t) llama_n_ctx(ctx_tgt)));
slot.smpl.reset(common_sampler_init(model_tgt, task.params.sampling));
} catch (std::exception & e) {
std::string err_msg = std::string("Failed to initialize samplers: ") + e.what();
send_error(task, err_msg, ERROR_TYPE_INVALID_REQUEST);
@@ -4148,7 +4147,6 @@ std::unique_ptr<server_res_generator> server_routes::handle_completions_impl(
task.params = server_schema::eval_llama_cmpl_schema(
ctx_server.vocab,
params,
meta->slot_n_ctx,
meta->logit_bias_eog,
data);
+4 -14
View File
@@ -124,8 +124,8 @@ std::vector<std::unique_ptr<field>> make_llama_cmpl_schema(const common_params &
->set_desc("Dynamic temperature exponent, controls how entropy maps to temperature"));
add((new field_num("repeat_last_n", params.sampling.penalty_last_n))
->set_hard_limits(-1, INT32_MAX)
->set_desc("Last n tokens to consider for penalizing repetition (0 = disabled, -1 = ctx-size)"));
->set_hard_limits(0, INT32_MAX)
->set_desc("Last n tokens to consider for penalizing repetition (0 = disabled)"));
add((new field_num("repeat_penalty", params.sampling.penalty_repeat))
->set_desc("Control the repetition of token sequences in the generated text (1.0 = disabled)"));
@@ -151,8 +151,8 @@ std::vector<std::unique_ptr<field>> make_llama_cmpl_schema(const common_params &
->set_desc("Tokens that extend repetition beyond this length receive exponentially increasing penalty: multiplier * base ^ (sequence_length - allowed_length)"));
add((new field_num("dry_penalty_last_n", params.sampling.dry_penalty_last_n))
->set_hard_limits(-1, INT32_MAX)
->set_desc("How many tokens to scan for repetitions (0 = disabled, -1 = context size)"));
->set_hard_limits(0, INT32_MAX)
->set_desc("How many tokens to scan for repetitions (0 = disabled)"));
add((new field_num("mirostat", params.sampling.mirostat))
->set_limits(0, 2)
@@ -515,7 +515,6 @@ std::vector<std::unique_ptr<field>> make_llama_cmpl_schema(const common_params &
task_params eval_llama_cmpl_schema(
const llama_vocab * vocab,
const common_params & params_base,
const int n_ctx_slot,
const std::vector<llama_logit_bias> & logit_bias_eog,
const json & data) {
task_params params;
@@ -549,15 +548,6 @@ task_params eval_llama_cmpl_schema(
// post-processing
{
if (params.sampling.penalty_last_n == -1) {
// note: should be the slot's context and not the full context, but it's ok
params.sampling.penalty_last_n = n_ctx_slot;
}
if (params.sampling.dry_penalty_last_n == -1) {
params.sampling.dry_penalty_last_n = n_ctx_slot;
}
// if "reasoning_format" is not provided, its handler will not be called, we will need to handle it here
auto reasoning_format = params.chat_parser_params.reasoning_format;
params.chat_parser_params.reasoning_in_content = params.stream && (reasoning_format == COMMON_REASONING_FORMAT_DEEPSEEK_LEGACY);
-1
View File
@@ -98,7 +98,6 @@ std::vector<std::unique_ptr<field>> make_llama_cmpl_schema(
task_params eval_llama_cmpl_schema(
const llama_vocab * vocab,
const common_params & params_base,
const int n_ctx_slot,
const std::vector<llama_logit_bias> & logit_bias_eog,
const json & data);