Morph MORPH
SIGN IN SIGN UP
dhruvhead / vllm-metal UNCLAIMED
0 0 2 Python

COMMITS

main
August 28, 2026
L
Optimize pure LoRA prefill routing (#647)
Lik Xun Yuan (Lx) committed
14705ad
August 27, 2026
M
[Bugfix] Normalize disabled top_k rows to vocab_size in mixed sampling batches (#646) (#648)
Mihal Dimo committed
813e738
August 26, 2026
C
Bump vLLM to 0.28.0 (#645)
Chao-Ju Chen committed
db4d7c5
M
[Spec Decode] Chunk cold draft-KV ingest to bound first-propose dispatch stall (vllm-project#482 direction 3) (#643)
Mihal Dimo committed
24c9c5b
August 25, 2026
C
[Test] Hoist buffer default test imports (#642)
Chao-Ju Chen committed
65ecad6
F
Keep the hybrid APC parity harness bootable on 16 GB hosts (#639)
fdaluiso committed
6ef8313
C
[Perf] Conditionally restore the MLX buffer memory default (#637)
Chao-Ju Chen committed
41f3ed3
August 24, 2026
L
Fix LoRA adapter cache eviction (#641)
Lik Xun Yuan (Lx) committed
790b09f
August 22, 2026
M
[Spec Decode] Skip re-ingesting accepted draft tokens whose KV the lookahead already wrote (#482 direction 2) (#636)
Mihal Dimo committed
211a1e4
L
Support remote GGUF references (#635)
Lik Xun Yuan (Lx) committed
3c9bbdc
August 21, 2026
R
Write align-mode GDN state rows in place with a Metal kernel (#634)
Ranran committed
acce614
L
Use loopback for default UniProc rendezvous (#633)
Lik Xun Yuan (Lx) committed
aa6d961
August 20, 2026
R
Remove per-layer GDN prefill materialization barriers (#632)
Ranran committed
67100ba
M
[Spec Decode] Register draft-model KV as a scheduler-managed cache group (#630)
Mihal Dimo committed
4d48fbc
August 19, 2026
C
[Bugfix] Pin torch sampling to CPU to stop random token corruption (#629)
Chao-Ju Chen committed
bd32be8
A
Correctly log dense KV cache size from previously allocated array nbytes (#627)
AstroWebDev committed
f4651f7
R
Add M5 NAX prefill kernel for paged GQA attention (#623)
Ranran committed
1c171b8
L
Support Hugging Face config sources for local GGUF weights (#624)
Lik Xun Yuan (Lx) committed
489da60
August 18, 2026
C
Compile stateless MLP blocks for decode (#621)
Chao-Ju Chen committed
4edfe6a
August 17, 2026
R
Defer the GDN recurrent decode state scatter (#620)
Ranran committed
150dd32
August 16, 2026
L
Trim weak platform block-size tests (#613)
Lik Xun Yuan (Lx) committed
a8b7e75
L
Add BGE-M3 encoder pooling (#612)
Lik Xun Yuan (Lx) committed
ef3ec8c
August 15, 2026
C
Enable the decode pipeline by default (#611)
Chao-Ju Chen committed
083f581
L
Add encoder pooling backend foundation (#608)
Lik Xun Yuan (Lx) committed
0fa5496
R
Document Qwen3.8 support and update status markers (#609)
Ranran committed
494fb3e
August 14, 2026
L
Refactor pooling behind a decoder backend (#606)
Lik Xun Yuan (Lx) committed
dce0517
L
Trim config contract test scaffolding (#604)
Lik Xun Yuan (Lx) committed
ca6059a
August 13, 2026
C
Run body-only forwards for intermediate prefill chunks (#592)
Chao-Ju Chen committed
0568788
C
Bump vLLM to 0.27.1 (#603)
Chao-Ju Chen committed
e35b529
L
Trim obsolete Metal platform capability glue (#602)
Lik Xun Yuan (Lx) committed
2cb7e4b