Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-19 01:05:09 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'upstream' into concedo_experimental

Browse Source 
# Conflicts:
#	.github/workflows/python-type-check.yml
#	AGENTS.md
#	CONTRIBUTING.md
#	examples/model-conversion/scripts/embedding/run-original-model.py
#	examples/model-conversion/scripts/utils/compare_tokens.py
#	examples/pydantic_models_to_grammar.py
#	ggml/src/ggml-rpc/ggml-rpc.cpp
#	pyrightconfig.json
#	scripts/compare-llama-bench.py
#	scripts/jinja/jinja-tester.py
#	scripts/server-bench.py
#	tests/test-grammar-integration.cpp
#	tests/test-grammar-parser.cpp
#	tests/test-llama-grammar.cpp
#	tests/test-tokenizer-random.py
#	tools/cli/README.md
#	tools/completion/README.md
#	tools/llama-bench/llama-bench.cpp
#	tools/server/README.md
This commit is contained in:
Concedo
2026-03-22 23:39:13 +08:00
parent 592dedee28 3306dbaef7
commit ef854f002e
18 changed files with 204 additions and 108 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-context.cpp
+11 -4
View File
@@ -1349,8 +1349,11 @@ int llama_context::encode(const llama_batch & batch_inp) {
const llama_seq_id seq_id = ubatch.seq_id_unq[s];
const int32_t seq_idx = ubatch.seq_idx[seq_id];
embd_seq_out[seq_id].resize(n_embd);
ggml_backend_tensor_get_async(backend_embd, t_embd, embd_seq_out[seq_id].data(), (n_embd*seq_idx)*sizeof(float), n_embd*sizeof(float));
// use n_embd_out (not n_embd_inp) - the pooled embedding has the model's
// output dimension, which differs from input dimension for deepstack models (e.g. qwen3vl)
const uint32_t n_embd_out = hparams.n_embd_out();
embd_seq_out[seq_id].resize(n_embd_out);
ggml_backend_tensor_get_async(backend_embd, t_embd, embd_seq_out[seq_id].data(), (n_embd_out*seq_idx)*sizeof(float), n_embd_out*sizeof(float));
}
} break;
case LLAMA_POOLING_TYPE_RANK:
@@ -1771,12 +1774,16 @@ int llama_context::decode(const llama_batch & batch_inp) {
// extract sequence embeddings (cleared before processing each batch)
auto & embd_seq_out = embd_seq;
// use n_embd_out (not n_embd_inp) - the pooled embedding has the model's
// output dimension, which differs from input dimension for deepstack models (e.g. qwen3vl)
const uint32_t n_embd_out = hparams.n_embd_out();
for (uint32_t s = 0; s < ubatch.n_seqs_unq; ++s) {
const llama_seq_id seq_id = ubatch.seq_id_unq[s];
const int32_t seq_idx = ubatch.seq_idx[seq_id];
embd_seq_out[seq_id].resize(n_embd);
ggml_backend_tensor_get_async(backend_embd, t_embd, embd_seq_out[seq_id].data(), (n_embd*seq_idx)*sizeof(float), n_embd*sizeof(float));
embd_seq_out[seq_id].resize(n_embd_out);
ggml_backend_tensor_get_async(backend_embd, t_embd, embd_seq_out[seq_id].data(), (n_embd_out*seq_idx)*sizeof(float), n_embd_out*sizeof(float));
}
} break;
case LLAMA_POOLING_TYPE_RANK: