Skip to content

grok 强兼: bridge openai/grok and xai-org/grok-1#83

Open
Jah-yee wants to merge 4 commits intoopenai:mainfrom
Jah-yee:main
Open

grok 强兼: bridge openai/grok and xai-org/grok-1#83
Jah-yee wants to merge 4 commits intoopenai:mainfrom
Jah-yee:main

Conversation

@Jah-yee
Copy link

@Jah-yee Jah-yee commented Mar 3, 2026

强兼 (Forceful Compatibility)

This PR bridges two projects that share the name grok:

  • openai/grok — grokking research (Power et al.)
  • xai-org/grok-1 — Grok-1 314B MoE LLM

What this adds

  • Bridge A: Grok-1-style architecture (MoE, RoPE, RMSNorm, gated GELU) in the OpenAI grokking framework → GrokOneTransformer, trainable on modular arithmetic to study whether MoE affects grokking.
  • Bridge B: OpenAI-style arithmetic evaluation in the Grok-1 inference pipeline.
  • Bridge C: TransformerConfig.to_grokking_config() for scaling Grok-1 config to grokking experiments.

Unified entry point: does_grok_grok.py (demo, train, eval-grok1).

All original behavior is preserved; bridge code is additive and marked with 强兼 comments.

Ref: Jah-yee/grok

Made with Cursor

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant