furyhawk/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-07-21 02:05:35 +00:00
Code Issues Packages Projects Releases Wiki Activity
4,735 Commits 952 Branches 1 Tag
main
Commit Graph
6 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
usrlocalbenandGitHub 65891dcdb3 include DSA indexer state in kv/slot serializer (#2146) 2026-07-18 09:01:26 +03:00
usrlocalbenandGitHub 9647246458 fix: token "corruption" due to wrong RoPE type for GLM-DSA (#2099) 
The DSA lightning indexer hardcoded LLAMA_ROPE_TYPE_NEOX for indexer Q/K
positional encoding, but the GLM-5.2 model config.json explicitly sets
"indexer_rope_interleave": true, meaning the indexer uses interleaved
RoPE (LLAMA_ROPE_TYPE_NORM).

GLM-5.1 config.json also sets "rope_interleave": true.

The mismatch caused incorrect indexer scores, wrong top-k key selection,
and single-character token errors and other corruption that could grow
worse with context length.

Here the setup is changed to use the model's given rope_type instead of
hardcoding NEOX. This fixes GLM-5.2, GLM-5.1, and should be less brittle
for future GLM variants.
 2026-07-09 08:49:21 +03:00
usrlocalbenandGitHub a5e41bc210 fix: GLM-DSA regression from #2067 (server crash with --spec-type mtp) (#2071) 
- fix incorrect reshape in the small-batch (n_tokens <= 8) indexer path
- anchor inp_dsa_sink in the graph with ggml_build_forward_expand
 2026-07-03 08:24:06 +02:00
usrlocalbenandGitHub c1c3421462 Fix incorrect --amb n_max_head fitting (#1375) 
kv_f32_size should be fit to --amb by number of divisions, not heads per
division.

Regression in b85a2a5
 2026-03-07 09:01:14 +01:00
usrlocalbenandGitHub e5622a2e91 Fix Phi-3, Phi-4 (#1226) 
* fix phi3 tensor setup

* avoid SWA for Phi-4
 2026-02-04 11:57:50 +02:00
usrlocalbenandGitHub f98b1befdb remove curious assertions (#705) 
This assertion can hit during prefill as MLA/KV tensors grow,
e.g. Kimi K2 n_ctx >= 32768.
 2025-08-19 14:41:29 +03:00