🤝 Open to Collab

Steffen Röcker PRO

sroecker

·

https://x.com/sroecker

AI & ML interests

Local models

Recent Activity

liked a Space about 1 hour ago

multimodalart/jev-decision-index

liked a model about 12 hours ago

cua-ai/cua-s1-4b-0.2

liked a model about 12 hours ago

RedHatAI/Qwen3.8-27B-speculator.dspark

View all activity

Organizations

upvoted an article 21 days ago

Article

Training a coding model to paint watercolours with TRL and OpenEnv

sergiopaniego

•

21 days ago

• 71

upvoted a paper 24 days ago

Puro-2B: Poor Lab's Qwen2-1.5B Trained on RTX 5090 within $5090

Paper • 2608.27370 • Published 28 days ago • 40

upvoted a collection 24 days ago

Puro-2B

普罗-2B • 11 items • Updated 27 days ago • 12

upvoted a collection about 1 month ago

🌟 Antares

Agentic Models for Vulnerability Localization • 2 items • Updated Jul 21 • 62

upvoted an article about 1 month ago

Article

Training a coding agent using the OpenCode harness in remote HF sandboxes with TRL and OpenEnv

sergiopaniego

•

Aug 5

• 29

upvoted 4 collections 4 months ago

🧬 Carbon

Carbon 500M, 3B, 8B genomic models and GGUF variants for llama.cpp • 7 items • Updated Jun 2 • 46

NuExtract3

12 items • Updated May 26 • 16

GSQ

GSQ: Highly-Accurate Low-Precision Scalar Quantization for LLMs via Gumbel-Softmax Sampling, https://huggingface.co/papers/2604.18556 • 10 items • Updated Aug 22 • 14

Fixed Chat Templates for Qwen 3.5 & 3.6

Rewritten Jinja templates fixing 5 bugs in official Qwen 3.5/3.6. Works in LM Studio, llama.cpp, MLX, vLLM. • 1 item • Updated Apr 30 • 11

upvoted a paper 4 months ago

Achieving Gold-Medal-Level Olympiad Reasoning via Simple and Unified Scaling

Paper • 2605.13301 • Published May 13 • 166

upvoted a collection 4 months ago

SpecDrift

Models released as a part of Attention-Drift Paper, trained for deployment on production • 2 items • Updated May 10 • 2

upvoted 3 collections 5 months ago

Gemma 4 Assistant GGUF

Gemma 4 MTP assistant drafters as GGUF (F16/Q8_0/Q5_K_M/Q4_K_M/Q4_K_S). Speculative-decoding heads for the atomic-llama-cpp-turboquant fork. • 4 items • Updated May 7 • 15

8GB VRAM Local LLMs - Practitioner Tested

11 models on RTX 4060 Ti 8GB. speed king: Llama 1B (228 tok/s). quality king: Gemma 4 E4B (6/6). agentic king: gpt-oss-20b (10/10). • 21 items • Updated May 21 • 7

Granite 4.1 Language Models

Efficient language models for multilingual generation, coding, RAG, and AI assistant workflows. • 15 items • Updated 10 days ago • 70

upvoted an article 5 months ago

Article

Granite 4.1 LLMs: How They’re Built

ibm-granite

•

Apr 29

• 93

upvoted 5 collections 5 months ago

APEX Quants (GGUF)

MoE models quantized with the APEX Quantization technique ( https://github.com/mudler/apex-quant ) • 48 items • Updated 13 days ago • 149

1930 Coder

Fine-tuning the Talkie 13B 1930 model on agentic trajectories • 4 items • Updated May 5 • 4

Qwen3.6

4 items • Updated Apr 22 • 501

Laguna XS.2

Designed for agentic coding and long-horizon work on a local machine. Apache 2.0. • 5 items • Updated Jun 29 • 31

privacy-filter

OpenAI's privacy-filter fine0tuned models • 12 items • Updated Jul 13 • 12