matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-11 08:17:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama: update to commit 71e90e88 (#10192)

Browse Source
This commit is contained in:
Jeffrey Morgan
2025-04-16 18:14:01 -04:00
committed by GitHub
parent 369de832cd
commit 943464ccb8
160 changed files with 42219 additions and 33080 deletions
Download Patch File Download Diff File Expand all files Collapse all files

71
llama/patches/0003-embeddings.patch
View File

@@ -1,52 +1,43 @@
From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
From: Michael Yang <mxyng@pm.me>
Date: Mon, 16 Sep 2024 15:53:14 -0700
From: jmorganca <jmorganca@gmail.com>
Date: Tue, 8 Apr 2025 15:28:34 -0700
Subject: [PATCH] embeddings
allow a loaded model in llama.cpp to be used for
both embeddings and causal attention text generation
instead of forcing one or the error
---
src/llama-context.cpp | 2 +-
src/llama.cpp | 6 ++++--
2 files changed, 5 insertions(+), 3 deletions(-)
src/llama-context.cpp | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/src/llama-context.cpp b/src/llama-context.cpp
index 671d2a81..47e79ed4 100644
index 4735e98e..65135172 100644
--- a/src/llama-context.cpp
+++ b/src/llama-context.cpp
@@ -479,7 +479,7 @@ size_t llama_output_reserve(struct llama_context & lctx, size_t n_outputs) {
@@ -1232,7 +1232,7 @@ int llama_context::decode(llama_batch & inp_batch) {
int64_t n_outputs_all = 0;
// count outputs
- if (batch.logits && !embd_pooled) {
+ if (batch.logits) {
for (uint32_t i = 0; i < n_tokens_all; ++i) {
n_outputs_all += batch.logits[i] != 0;
}
@@ -1344,7 +1344,7 @@ int llama_context::decode(llama_batch & inp_batch) {
// ggml_graph_dump_dot(gf, NULL, "llama.dot");
//}
- auto * t_logits = cparams.embeddings ? nullptr : res->get_logits();
+ auto * t_logits = cparams.causal_attn ? res->get_logits() : nullptr;
auto * t_embd = cparams.embeddings ? res->get_embd() : nullptr;
if (t_embd && res->get_embd_pooled()) {
@@ -1488,7 +1488,7 @@ int32_t llama_context::output_reserve(int32_t n_outputs) {
const auto n_embd = hparams.n_embd;
// TODO: use a per-batch flag for logits presence instead
- const bool has_logits = !cparams.embeddings;
+ const bool has_logits = cparams.causal_attn;
const bool has_embd = cparams.embeddings && (cparams.pooling_type == LLAMA_POOLING_TYPE_NONE);
- bool has_logits = !cparams.embeddings;
+ bool has_logits = cparams.causal_attn;
bool has_embd = cparams.embeddings && (cparams.pooling_type == LLAMA_POOLING_TYPE_NONE);
const size_t logits_size = has_logits ? n_vocab*n_outputs_max : 0;
diff --git a/src/llama.cpp b/src/llama.cpp
index 607f2786..ac85bfed 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -8652,7 +8652,6 @@ static int llama_decode_impl(
res = nullptr;
embd = nullptr;
} else if (cparams.embeddings) {
- res = nullptr; // do not extract logits for embedding case
embd = nullptr;
for (int i = ggml_graph_n_nodes(gf) - 1; i >= 0; --i) {
if (strcmp(ggml_graph_node(gf, i)->name, "result_embd_pooled") == 0) {
@@ -8660,12 +8659,15 @@ static int llama_decode_impl(
break;
}
}
- GGML_ASSERT(embd != nullptr && "missing embeddings tensor");
} else {
embd = nullptr; // do not extract embeddings when not needed
GGML_ASSERT(strcmp(res->name, "result_output") == 0 && "missing result_output tensor");
}
+ if (!cparams.causal_attn) {
+ res = nullptr; // do not extract logits when not needed
+ }
+
// LLAMA_LOG_INFO("graph build time: %.3f ms (%d nodes, %d leafs)\n", (ggml_time_us() - t_start_us)/1000.0, gf->n_nodes, gf->n_leafs);
ggml_backend_sched_alloc_graph(lctx.sched.get(), gf);
// TODO: hacky enc-dec support