docs: refresh README — pegainfer.org links, blog list, GLM-5.2 row - #868
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 36ad5ee61e
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| | [Qwen3.5-4B](https://huggingface.co/Qwen/Qwen3.5-4B) / [9B](https://huggingface.co/Qwen/Qwen3.5-9B) / [27B](https://huggingface.co/Qwen/Qwen3.5-27B) | Hybrid Gated DeltaNet + full attention | 4B / 9B / 27B | Text-only BF16, greedy + sampling, feature-gated, `--features qwen35` (build-time Triton) | | ||
| | [DeepSeek-V2-Lite](https://huggingface.co/deepseek-ai/DeepSeek-V2-Lite) | MoE + EP | 15.7B total / 2.4B active | Feature-gated, `--features deepseek-v2-lite`, 2-GPU EP2 correctness path | | ||
| | [Kimi-K2-Instruct](https://huggingface.co/moonshotai/Kimi-K2-Instruct) | MLA + MoE + Marlin INT4 | 1T total / 32B active | Feature-gated, `--features kimi-k2`, 8-GPU EP path | | ||
| | [GLM-5.2](https://huggingface.co/zai-org/GLM-5.2-FP8) | Sparse MLA (DSA) + MoE + native MTP | FP8, ~704 GB | Feature-gated, `--features glm52`, Blackwell-only; EP decode, TP4 prefill, P/D disaggregation — see the [GLM-5.2 guide](https://pegainfer.org/models/glm52/) | |
There was a problem hiding this comment.
For readers comparing model sizes, this row places the checkpoint's quantization and approximate storage footprint (FP8, ~704 GB) under a Params heading, unlike every other row, which reports total and active parameter counts. This makes the public supported-model table factually inconsistent and can mislead capacity comparisons; report GLM-5.2's total/active parameter counts here, or rename and consistently repurpose the column.
Useful? React with 👍 / 👎.
- Point the docs badge and every open-infer.org link at pegainfer.org, using the canonical /blog/pegainfer-010/ slug. - Expand the docs paragraph into a list of the five blog posts. - Trim the API section to the endpoints and SDK base URL; the field reference lives on the website. - Add GLM-5.2 to the supported-models table, linking the model guide. - Correct the PEGAINFER_TILELANG_PYTHON row (feeds k3 codegen now, not glm52). - Drop the star-history section. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Signed-off-by: xiaguan <751080330@qq.com>
36ad5ee to
3c70847
Compare
1 participant
README refresh after the domain rename and the GLM-5.2 docs page landing:
open-infer.orglinks now point atpegainfer.org(old blog slug canonicalized to/blog/pegainfer-010/). All links verified live (200).--features glm52, Blackwell-only, EP decode / TP4 prefill / P/D) linking the new GLM-5.2 guide.PEGAINFER_TILELANG_PYTHONdescription was stale — it feedsk3build-time codegen now, not the glm52 sm_90a path.Architecture diagram intentionally untouched.
🤖 Generated with Claude Code