Skip to content

Navigation Menu

Appearance settings

View all features
- BY COMPANY SIZE
  Enterprises
  Small and medium teams
  Startups
  Nonprofits
- BY USE CASE
  App Modernization
  DevSecOps
  DevOps
  CI/CD
  View all use cases
- BY INDUSTRY
  Healthcare
  Financial services
  Manufacturing
  Government
  View all industries
View all solutions
- EXPLORE BY TOPIC
  AI
  Software Development
  DevOps
  Security
  View all topics
- EXPLORE BY TYPE
  Customer stories
  Events & webinars
  Ebooks & reports
  Business insights
  GitHub Skills
- SUPPORT & SERVICES
  Documentation
  Customer support
  Community forum
  Trust center
  Partners
View all resources
- COMMUNITY
  GitHub SponsorsFund open source developers
- PROGRAMS
  Security Lab
  Maintainer Community
  Accelerator
  GitHub Stars
  Archive Program
- REPOSITORIES
  Topics
  Trending
  Collections
- ENTERPRISE SOLUTIONS
  Enterprise platformAI-powered developer platform
- AVAILABLE ADD-ONS
  GitHub Advanced SecurityEnterprise-grade security features
  Copilot for BusinessEnterprise-grade AI features
  Premium SupportEnterprise-grade 24/7 support
Pricing

Search code, repositories, users, issues, pull requests...

Search

Clear

Search syntax tips

Provide feedback

We read every piece of feedback, and take your input very seriously.

Include my email address so I can be contacted

Saved searches

Use saved searches to filter your results more quickly

Name

Query

To see all available qualifiers, see our documentation.

Appearance settings

You signed in with another tab or window. Reload to refresh your session. You signed out in another tab or window. Reload to refresh your session. You switched accounts on another tab or window. Reload to refresh your session.

Dismiss alert

gpu-mode / paris-hackathon-2026-inference Public

Notifications You must be signed in to change notification settings
Fork 8
Star 1

Code
Issues
Pull requests 1
Actions
Projects
Security and quality
Insights

Additional navigation options

Code
Issues
Pull requests
Actions
Projects
Security and quality
Insights

wip#1

Open

yachty66 wants to merge 24 commits into

gpu-mode:maingpu-mode/paris-hackathon-2026-inference:mainfrom

FelixMul:maxhFelixMul/Inference_Engine:maxhCopy head branch name to clipboard

Conversation Commits24 (24)Checks Files changed

Open

wip#1
yachty66 wants to merge 24 commits into
gpu-mode:mainfrom
FelixMul:maxh

Commits

Commits on Apr 9, 2026

Add initial inference server
yachty66
committed
Add custom inference engine (GatedDeltaNet + GQA + MoE)
yachty66
committed
Fix q_norm/k_norm: use RMSNorm instead of LayerNorm (no bias)
yachty66
committed
Fix linear attn einsum: keep val-sub-head dim in outer product
yachty66
committed
Fix dtype mismatch: cast softmax output back to bfloat16
yachty66
committed
Fix generate: return [1, num_generated] via stack not cat
yachty66
committed
Fix causal conv1d (manual left-pad) and add Q normalization
yachty66
committed
Fix conv_buf: save from padded input not raw qkv_t
yachty66
committed
Add debug comparison script (HF vs our model)
yachty66
committed
Add layer-by-layer debug comparison script
yachty66
committed
Add source inspection script
yachty66
committed
Add source2 inspection script
yachty66
committed
Fix all bugs found from source: RMSNorm formula, conv silu, Q/K expand, sigmoid gate, per-head gate split
yachty66
committed
Add layer 0 sublayer debug script
yachty66
committed
Add GatedDeltaNet internals debug script
yachty66
committed
Fix syntax error in debug_deltanet.py
yachty66
committed
Fix delta rule: decay state first, then compute delta from decayed state
yachty66
committed
Fix debug script delta rule order to match fix
yachty66
committed
Dump HF chunk_gated_delta_rule torch fallback source

yachty66
and
claude
committed
Fix delta rule: scale query by 1/sqrt(d_k) before recurrence

yachty66
and
claude
committed
Stop generation on <|im_end|> as well as <|endoftext|>

yachty66
and
claude
committed
Use FLA chunk kernel for prefill + data-parallel across all GPUs

yachty66
and
claude
committed
Cast FLA kernel outputs back to input dtype

yachty66
and
claude
committed
Load model replicas across GPUs in parallel

yachty66
and
claude
committed

Footer

© 2026 GitHub, Inc.

Footer navigation

Terms
Privacy
Security
Status
Community
Docs
Contact

You can’t perform that action at this time.