#9797th of 105 in Model serving
clm
Contrastive model that scores candidate actions or answers for a given state
- Stars
- 3.2k
- License
- Apache-2.0
- Last commit
- Oct 2026
- Language
- Python
940 on 24 Sept 20263,212 on 19 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
CLM serves CLM-8B, a contrastive state-action model. It embeds a state and its candidates with a Qwen3-8B encoder served by vLLM, then scores them through 20M-parameter projection heads. A FastAPI server exposes typed questions (noul, choice, score) and a rank endpoint, plus a web playground. The repo also has a fine-tuning script and an N-candidate selection evaluation for verifier use.
Who it is for: Engineers building agent verifiers, routers or classifiers on embeddings
Strengths
- State and action embeddings are cached separately, so repeated candidates cost a dot product
- Returns probability distributions for typed questions via /v1/systemone and /v1/rank
- Ships a fine-tuning script, a candidate-selection eval and a playground UI on port 8700
- Code and CLM-8B weights are both Apache-2.0
Weaknesses
- Needs a separate vLLM Qwen3-8B pooling server on a GPU as the encoder
- States over 2048 tokens are truncated unless both server limits are raised
- Selects among supplied candidates only; it does not generate text
- Comparisons against Jev come from the authors' own blog and evaluation
What it needs
- GPU required
- Needs vLLM, Qwen/Qwen3-8B, PyTorch
- Models: CLM-8B (Qwen3-8B backbone), custom projection head checkpoints (.pt)
- port 8700
Known problems
Open now, reported in the last 90 days
- installation on mac ↗ (opens in a new tab)2 reactions
- Vector-cache slot leak on concurrent misses (measured numbers) ↗ (opens in a new tab)1 reactions
Titles are the reporters' own words, taken from the project's GitHub issues and ranked by reactions. We have not tried to reproduce any of them.
Also in Model serving
See all 105| Rank | Project | Score |
|---|---|---|
| 1 | llama.cppC/C++ inference engine for GGUF LLMs and VLMs with an OpenAI-compatible server | 94 out of 100 |
| 2 | OllamaRuns open models locally through a CLI and REST API | 91 out of 100 |
| 3 | LocalAISelf-hosted model server with OpenAI, Anthropic and ElevenLabs-compatible APIs | 91 out of 100 |
| #4 | strataDesktop and terminal app that runs GGUF models locally with OpenAI-compatible APIs | 89 out of 100 |
| #5 | modularMojo language and MAX inference framework with an OpenAI-compatible server | 89 out of 100 |