Fix a bug in tying OPT embeddings #1
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
orangetin
referenced
this pull request
in togethercomputer/vllm-ttgi
Sep 14, 2023
add rope scaling as a cli arg so openai server can load rope scaled models
xiangyuT
pushed a commit
to xiangyuT/vllm
that referenced
this pull request
Oct 18, 2023
bigPYJ1151
added a commit
to bigPYJ1151/vllm
that referenced
this pull request
Oct 30, 2023
Fix key cache block shape.
l1cacheDell
pushed a commit
to CaspianFang/vllm
that referenced
this pull request
Nov 15, 2023
hongxiayang
referenced
this pull request
in hongxiayang/vllm
Feb 13, 2024
ilya-lavrenov
referenced
this pull request
in ilya-lavrenov/vllm
Feb 19, 2024
Deterministic OpenVINO inference
daniel-geon-park
added a commit
to gmlwns2000/vllm-timber
that referenced
this pull request
Apr 15, 2024
mzusman
pushed a commit
to mzusman/vllm
that referenced
this pull request
Apr 16, 2024
BA-78554: Jurassic 2.5 * worked on jurasic2.5 configuration file, updated jurassic2_5 modeling file to support alternating experts/attn layers * finished working the forward pass of jurassic3.py * finished working the forward pass of jurassic3.py * finished working the forward pass of jurassic3.py * jurassic_3 modeling file works, uses dummy weights initialized by "dummy" flag. Tokenizer raises issues, for now copying the mixtral tokenizer * changed default tokenizer vocab values, loading of custom .pt weight files works. * removed notebook * merging master to jurassic-2.5 to reset head * Merge branch 'master' into jurassic-2.5 * align to master Approved-by: Tomer Asida Approved-by: Mor Zusman
Bellk17
added a commit
to Bellk17/vllm
that referenced
this pull request
May 10, 2024
Triton compilation fix
This was referenced Jul 5, 2024
1 task
1 task
1 task
1 task
1 task
1 task
1 task
bbartels
pushed a commit
to bbartels/vllm
that referenced
this pull request
Aug 14, 2025
…ermerge feat:trace v1
1 task
1 task
16 tasks
1 task
1 task
4 tasks
1 task
yuz207
referenced
this pull request
in IluvatarLabs/vllm
Sep 27, 2025
Bug #1 (CRITICAL): Add missing begin() and stage() methods to KVWriteRouter - Flash attention backend calls router.begin() and router.stage() - KVWriteRouter only had write() and commit() methods - Added begin() to store slot_mapping and initialize shadow buffer - Added stage() to extract per-timestep slot and stage KV pairs - Without these, no tokens were being staged → 0% acceptance rate Bug #2 (MODERATE): Fix bonus token counting in accepted_lens - valid_sampled_token_ids includes [accepted_draft_tokens..., bonus_token] - Previous: len([bonus]) = 1, incorrectly counted as 1 accepted draft token - Fixed: Use max(0, len(seq) - 1) to exclude bonus token from count - Now correctly reports 0 accepted when only bonus token is present Files modified: - vllm/v1/kv_cache/write_router.py: Added begin() and stage() methods - vllm/v1/worker/gpu_model_runner.py: Fixed accepted_lens calculation
yuz207
referenced
this pull request
in IluvatarLabs/vllm
Sep 27, 2025
Bug #1: EAGLE tree proposal returned zeros for draft_logprobs - Root cause: When using topk for tree branching, code set draft_logp_list=None, then created zeros tensor as fallback (lines 850-851) - Fix: Compute actual log-probs from logits using log_softmax + gather - Applied at 2 locations: root level (lines 698-704) and tree levels (lines 839-846) Bug #2: Added diagnostic logging in rejection sampler - Log draft_p (nonzero) min/med/max to detect zeros - Log p_target min/med/max to detect degenerate softmax - Helps identify if target logits are masked/filtered before sampling Expected results after fix: - draft_logp: -3.2/-1.6/-0.0 (real log-probs, all ≤ 0) instead of 0/0/0 - p_target: 1e-6/1e-3/0.7 (realistic distribution) instead of 1/1/1 - Acceptance rate: 30-70% instead of 0% Files changed: - vllm/v1/spec_decode/eagle.py: Fix draft_logp computation - vllm/v1/sample/rejection_sampler.py: Add sanity logging
yuz207
referenced
this pull request
in IluvatarLabs/vllm
Sep 30, 2025
CRITICAL FIX: tau_d was reading draft_temperature (0.05) instead of target temperature from sampling_metadata (1.0). This caused: - tau_q = 0.05 + 0.3 = 0.35 (before) - Logit gap = 10/0.35 = 28.6 → exp(-28.6) ≈ 0 (underflow!) - q collapses to 0.98-1.0 After fix: - tau_d = 1.0 (from sampling_metadata.temperature) - tau_q = 1.0 + 0.3 = 1.3 - Logit gap = 10/1.3 = 7.7 → exp(-7.7) = 0.00045 (survives!) - q should be in [0.5, 0.8] range Changes: - propose(): Store sampling_metadata as self._current_sampling_metadata - _sample_draft_tokens(): Read tau_d from sampling_metadata, not opt_config
1 task
Pradyun92
pushed a commit
to Pradyun92/vllm
that referenced
this pull request
Oct 20, 2025
Enhanced documentation for plugin patches: 1. Patch vllm-project#1 (Usage Tracking Helper): - Clarified as OPTIONAL (has fallback in harmony streaming patch) - Changed from "REQUIRED" to "OPTIONAL" - Explained fallback mechanism in patched_stream_method.py - Marked as upstreamable (minor utility addition) 2. Patch vllm-project#3 (Harmony Token-by-Token Streaming): - Added detailed speculative decoding context - Explained Eagle draft model generates 5-10 tokens per step - Documented specific failures with batch processing: * Tool calling broken * Multi-channel content lost * Token truncation during channel transitions - Added before/after code examples - Linked to PR vllm-project#26291 (Eagle3 Multi-Channel Streaming Fix) - Documented upstream status and removal plan Key insight: This patch exists because Eagle speculative decoding returns multiple tokens per step, and upstream's batch processing can't handle per-token channel switching. Signed-off-by: Pradyun Ramadorai <[email protected]>
1 task
1 task
1 task
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR fixes a bug in supporting OPT-350m/OPT-6.7b/OPT-13b and OPT-IML models.
The bug happened because our model code didn't include some methods that were required to tie the input and output embeddings.