ispobock
Follow
Pinned Loading
402 contributions in the last year
Less
No contributions.
Low contributions.
Medium-low contributions.
Medium-high contributions.
High contributions.
More
Activity overview
Contributed to
sgl-project/sglang,
InternLM/lmdeploy,
ispobock/test-workflow
and 17 other
repositories
Loading
Contribution activity
March 2025
Created 17 commits in 1 repository
Created 1 repository
-
ispobock/EAGLE
Python
This contribution was made on Mar 19
Created a pull request in sgl-project/sglang that received 4 comments
Fix EAGLE3 for llama3.3 70b
Motivation The EAGLE3-LLaMA3.3-Instruct-70B draft model has a different hidden size with the target model. It should be handled in fc layer. Ref: h…
+10
−1
lines changed
•
4
comments
Opened 15 other pull requests in 1 repository
sgl-project/sglang
14
merged
1
closed
-
update torch compile doc
This contribution was made on Mar 29
-
Update doc for DeepSeek-V3-0324
This contribution was made on Mar 27
-
Add EAGLE mtbench benchmark script
This contribution was made on Mar 22
-
Update doc for MTP and DP attention
This contribution was made on Mar 20
-
Add deepseek v2 torch compile pr test
This contribution was made on Mar 18
-
Revert "feat: update grouped_topk to support softmax and sigmoid"
This contribution was made on Mar 17
-
Update MTP doc
This contribution was made on Mar 11
-
Update bench speculative script
This contribution was made on Mar 9
-
Fix eagle hang issue for max_new_tokens=1
This contribution was made on Mar 7
-
Remove prefill one req limit for eagle
This contribution was made on Mar 6
-
Fix triton kernel illegal memory issue for eagle
This contribution was made on Mar 5
-
Update nextn ci test
This contribution was made on Mar 4
-
Share target model embed and head weights for nextn
This contribution was made on Mar 3
-
Add accuracy test for TP torch compile
This contribution was made on Mar 2
-
Fix all gather torch compile
This contribution was made on Mar 2
Reviewed 16 pull requests in 1 repository
sgl-project/sglang
16 pull requests
-
Feat/support encoder model (like bert)
This contribution was made on Mar 30
-
feat(adaptive speculative): Implement Multi-Armed Bandit (MAB) for adaptive speculative decoding
This contribution was made on Mar 26
-
Support FA3 as Attention backend by using
--attention-backend fa3This contribution was made on Mar 24 -
add dsv3 int8 test
This contribution was made on Mar 24
-
[Kernel] Support BF16 awq_dequant
This contribution was made on Mar 18
-
Support Online Quantization for W8A8
This contribution was made on Mar 17
-
Unit test for Hierarchical Caching
This contribution was made on Mar 17
-
[Feature] Support EAGLE 3
This contribution was made on Mar 12
-
Add INT8 support MTP NextN function
This contribution was made on Mar 12
-
refactor: move image processors to separate files
This contribution was made on Mar 11
-
Add A800 tuning configs support DeepSeek V3/R1 BF16 and INT8(block-wise)
This contribution was made on Mar 9
-
add INT8 example into dsv3 README
This contribution was made on Mar 6
-
Remove prefill-only-one-req
This contribution was made on Mar 6
-
[Bug fixed] fixed the crash when enable the dp-attention on the single card
This contribution was made on Mar 5
-
[Eagle] Refactor eagle speculative decoding
This contribution was made on Mar 5
-
add deepgemm into sgl-kernel
This contribution was made on Mar 1
Answered 1 discussion in 1 repository
sgl-project/sglang
sgl-project/sglang
-
about MLA kv cache
This contribution was made on Mar 10