Skip to content

llama : support multiple classifier outputs and labels #13940

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 12 commits into from
Jun 6, 2025
Merged

llama : support multiple classifier outputs and labels #13940

Show file tree
Hide file tree
Changes from 10 commits
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
a1b1018
add multiple classifier outputs and labels support
CISC May 31, 2025
6ef43ba
make public
CISC May 31, 2025
fa61273
show multiple rankings and associated labels
CISC May 31, 2025
38ece05
move labels to llama_model
CISC Jun 1, 2025
76cf024
Merge branch 'master' into cisc/multiple-classifier-outputs
CISC Jun 1, 2025
3a52f4c
update n_cls_out for any arch with labels
CISC Jun 2, 2025
41049e6
be more specific about behaviour
CISC Jun 2, 2025
5b6bf2f
move string array functionality into model-loader
CISC Jun 2, 2025
9aa5d73
forgotten variable replacements
CISC Jun 2, 2025
6bba8ed
improved comment [no ci]
CISC Jun 3, 2025
7443156
logging and minor changes
CISC Jun 5, 2025
67c4cd2
forgot to remove cstring
CISC Jun 5, 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
20 changes: 18 additions & 2 deletions examples/embedding/embedding.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,7 @@
#include "llama.h"

#include <ctime>
#include <cstring>
#include <algorithm>

#if defined(_MSC_VER)
Expand Down Expand Up @@ -236,9 +237,24 @@ int main(int argc, char ** argv) {
LOG("\n");
}
} else if (pooling_type == LLAMA_POOLING_TYPE_RANK) {
const uint32_t n_cls_out = llama_model_n_cls_out(model);
std::vector<std::string> cls_out_labels;

for (uint32_t i = 0; i < n_cls_out; i++) {
const char * label = llama_model_cls_label(model, i);
const std::string label_i = label == nullptr || strlen(label) == 0 ? std::to_string(i) : label;
cls_out_labels.emplace_back(label_i);
}

for (int j = 0; j < n_embd_count; j++) {
// NOTE: if you change this log - update the tests in ci/run.sh
LOG("rerank score %d: %8.3f\n", j, emb[j * n_embd]);
for (uint32_t i = 0; i < n_cls_out; i++) {
// NOTE: if you change this log - update the tests in ci/run.sh
if (n_cls_out == 1) {
LOG("rerank score %d: %8.3f\n", j, emb[j * n_embd]);
} else {
LOG("rerank score %d: %8.3f [%s]\n", j, emb[j * n_embd + i], cls_out_labels[i].c_str());
}
}
}
} else {
// print the first part of the embeddings or for a single prompt, the full embedding
Expand Down
9 changes: 8 additions & 1 deletion include/llama.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -509,6 +509,13 @@ extern "C" {
// Get the model's RoPE frequency scaling factor
LLAMA_API float llama_model_rope_freq_scale_train(const struct llama_model * model);

// Returns the number of classifier outputs (only valid for classifier models)
// Undefined behavior for non-classifier models
LLAMA_API uint32_t llama_model_n_cls_out(const struct llama_model * model);

// Returns label of classifier output by index (<n_cls_out). Returns nullptr if no label provided
LLAMA_API const char * llama_model_cls_label(const struct llama_model * model, uint32_t i);

LLAMA_API enum llama_vocab_type llama_vocab_type(const struct llama_vocab * vocab);

LLAMA_API int32_t llama_vocab_n_tokens(const struct llama_vocab * vocab);
Expand Down Expand Up @@ -916,7 +923,7 @@ extern "C" {

// Get the embeddings for a sequence id
// Returns NULL if pooling_type is LLAMA_POOLING_TYPE_NONE
// when pooling_type == LLAMA_POOLING_TYPE_RANK, returns float[1] with the rank of the sequence
// when pooling_type == LLAMA_POOLING_TYPE_RANK, returns float[n_cls_out] with the rank(s) of the sequence
// otherwise: float[n_embd] (1-dimensional)
LLAMA_API float * llama_get_embeddings_seq(struct llama_context * ctx, llama_seq_id seq_id);

Expand Down
7 changes: 4 additions & 3 deletions src/llama-context.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -814,16 +814,17 @@ int llama_context::encode(llama_batch & inp_batch) {
} break;
case LLAMA_POOLING_TYPE_RANK:
{
// extract the rerank score - a single float per sequence
// extract the rerank score - n_cls_out floats per sequence
auto & embd_seq_out = embd_seq;
const uint32_t n_cls_out = hparams.n_cls_out;

for (uint32_t s = 0; s < ubatch.n_seqs; ++s) {
const llama_seq_id seq_id = ubatch.seq_id[s][0];
if (embd_seq_out.find(seq_id) != embd_seq_out.end()) {
continue;
}
embd_seq_out[seq_id].resize(1);
ggml_backend_tensor_get_async(backend_embd, t_embd, embd_seq_out[seq_id].data(), (seq_id)*sizeof(float), sizeof(float));
embd_seq_out[seq_id].resize(n_cls_out);
ggml_backend_tensor_get_async(backend_embd, t_embd, embd_seq_out[seq_id].data(), (n_cls_out*seq_id)*sizeof(float), n_cls_out*sizeof(float));
}
} break;
case LLAMA_POOLING_TYPE_UNSPECIFIED:
Expand Down
59 changes: 42 additions & 17 deletions src/llama-model-loader.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -288,61 +288,84 @@ namespace GGUFMeta {

template<typename T>
bool llama_model_loader::get_arr(const std::string & key, std::vector<T> & result, bool required) {
const int kid = gguf_find_key(meta.get(), key.c_str());
const gguf_context * ctx = meta.get();
const int kid = gguf_find_key(ctx, key.c_str());

if (kid < 0 || gguf_get_kv_type(meta.get(), kid) != GGUF_TYPE_ARRAY) {
if (kid < 0 || gguf_get_kv_type(ctx, kid) != GGUF_TYPE_ARRAY) {
if (required) {
throw std::runtime_error(format("array key not found in model: %s", key.c_str()));
}
return false;
}

struct GGUFMeta::ArrayInfo arr_info =
GGUFMeta::GKV<GGUFMeta::ArrayInfo>::get_kv(meta.get(), kid);
GGUFMeta::GKV<GGUFMeta::ArrayInfo>::get_kv(ctx, kid);

switch (arr_info.gt) {
case GGUF_TYPE_UINT32:
case GGUF_TYPE_INT32: GGML_ASSERT((std::is_same<T, int32_t>::value) ||
(std::is_same<T, uint32_t>::value)); break;
case GGUF_TYPE_FLOAT32: GGML_ASSERT((std::is_same<T, float>::value)); break;
case GGUF_TYPE_INT32: GGML_ASSERT((std::is_same<T, int32_t>::value) ||
(std::is_same<T, uint32_t>::value)); break;
case GGUF_TYPE_FLOAT32: GGML_ASSERT((std::is_same<T, float>::value)); break;
case GGUF_TYPE_STRING: GGML_ASSERT((std::is_same<T, std::string>::value)); break;
default:
throw std::runtime_error(format("%s is not a float32/uint32/int32 array", key.c_str()));
throw std::runtime_error(format("%s is not a string/float32/uint32/int32 array", key.c_str()));
}

result.resize(arr_info.length);
result.assign((const T*)arr_info.data, (const T *)arr_info.data + arr_info.length);
if constexpr (std::is_same<T, std::string>::value) {
const size_t n_items = gguf_get_arr_n(ctx, kid);
result.clear();

for (size_t i = 0; i < n_items; i++) {
const T value = gguf_get_arr_str(ctx, kid, i);
result.emplace_back(value);
}
} else {
result.resize(arr_info.length);
result.assign((const T*)arr_info.data, (const T *)arr_info.data + arr_info.length);
}

return true;
}

template<typename T, size_t N_MAX>
bool llama_model_loader::get_arr(const std::string & key, std::array<T, N_MAX> & result, bool required) {
const int kid = gguf_find_key(meta.get(), key.c_str());
const gguf_context * ctx = meta.get();
const int kid = gguf_find_key(ctx, key.c_str());

if (kid < 0 || gguf_get_kv_type(meta.get(), kid) != GGUF_TYPE_ARRAY) {
if (kid < 0 || gguf_get_kv_type(ctx, kid) != GGUF_TYPE_ARRAY) {
if (required) {
throw std::runtime_error(format("array key not found in model: %s", key.c_str()));
}
return false;
}

struct GGUFMeta::ArrayInfo arr_info =
GGUFMeta::GKV<GGUFMeta::ArrayInfo>::get_kv(meta.get(), kid);
GGUFMeta::GKV<GGUFMeta::ArrayInfo>::get_kv(ctx, kid);

switch (arr_info.gt) {
case GGUF_TYPE_UINT32:
case GGUF_TYPE_INT32: GGML_ASSERT((std::is_same<T, int32_t>::value) ||
(std::is_same<T, uint32_t>::value)); break;
case GGUF_TYPE_FLOAT32: GGML_ASSERT((std::is_same<T, float>::value)); break;
case GGUF_TYPE_INT32: GGML_ASSERT((std::is_same<T, int32_t>::value) ||
(std::is_same<T, uint32_t>::value)); break;
case GGUF_TYPE_FLOAT32: GGML_ASSERT((std::is_same<T, float>::value)); break;
case GGUF_TYPE_STRING: GGML_ASSERT((std::is_same<T, std::string>::value)); break;
default:
throw std::runtime_error(format("%s is not a float32/uint32/int32 array", key.c_str()));
throw std::runtime_error(format("%s is not a string/float32/uint32/int32 array", key.c_str()));
}

if (arr_info.length > N_MAX) {
throw std::runtime_error(format("array length %u for key %s exceeds max %u", (uint32_t) arr_info.length, key.c_str(), (uint32_t) N_MAX));
}

std::copy((const T*)arr_info.data, (const T *)arr_info.data + arr_info.length, result.begin());
if constexpr (std::is_same<T, std::string>::value) {
const size_t n_items = gguf_get_arr_n(ctx, kid);

for (size_t i = 0; i < n_items; i++) {
const T value = gguf_get_arr_str(ctx, kid, i);
result[i] = value;
}
} else {
std::copy((const T*)arr_info.data, (const T *)arr_info.data + arr_info.length, result.begin());
}

return true;
}
Expand All @@ -352,6 +375,8 @@ namespace GGUFMeta {
return get_arr(llm_kv(kid), result, required);
}

template bool llama_model_loader::get_arr<std::vector<std::string>>(enum llm_kv kid, std::vector<std::string> & result, bool required);

template<typename T>
bool llama_model_loader::get_key(const std::string & key, T & result, bool required) {
auto it = kv_overrides.find(key);
Expand Down
19 changes: 18 additions & 1 deletion src/llama-model.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -543,6 +543,12 @@ void llama_model::load_hparams(llama_model_loader & ml) {
uint32_t n_vocab = 0;
ml.get_key(LLM_KV_VOCAB_SIZE, n_vocab, false) || ml.get_arr_n(LLM_KV_TOKENIZER_LIST, n_vocab, false);

// for classifier models
ml.get_arr(LLM_KV_CLASSIFIER_OUTPUT_LABELS, classifier_labels, false);
if (!classifier_labels.empty()) {
hparams.n_cls_out = classifier_labels.size();
}

// arch-specific KVs
switch (arch) {
case LLM_ARCH_LLAMA:
Expand Down Expand Up @@ -686,7 +692,6 @@ void llama_model::load_hparams(llama_model_loader & ml) {
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
ml.get_key(LLM_KV_ATTENTION_CAUSAL, hparams.causal_attn);
ml.get_key(LLM_KV_POOLING_TYPE, hparams.pooling_type, false);
ml.get_arr_n(LLM_KV_CLASSIFIER_OUTPUT_LABELS, hparams.n_cls_out, false);

switch (hparams.n_layer) {
case 3:
Expand Down Expand Up @@ -13600,6 +13605,18 @@ int32_t llama_model_n_swa(const llama_model * model) {
return model->hparams.n_swa;
}

uint32_t llama_model_n_cls_out(const struct llama_model * model) {
return model->hparams.n_cls_out;
}

const char * llama_model_cls_label(const struct llama_model * model, uint32_t i) {
if (i < model->classifier_labels.size()) {
return model->classifier_labels[i].c_str();
}

return nullptr;
}

// deprecated
int32_t llama_n_ctx_train(const llama_model * model) {
return llama_model_n_ctx_train(model);
Expand Down
3 changes: 3 additions & 0 deletions src/llama-model.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -363,6 +363,9 @@ struct llama_model {
// for quantize-stats only
std::vector<std::pair<std::string, struct ggml_tensor *>> tensors_by_name;

// for classifier models
std::vector<std::string> classifier_labels;

int64_t t_load_us = 0;
int64_t t_start_us = 0;

Expand Down