AI Researcher @ delta : kitsune
-
nanbeige4.1
3B model that shouldn't be this good - crushes benchmarks through deep chain-of-thought reasoning
1,946 Pulls 1 Tag Updated 8 months ago
-
Mellum2-Thinking-A2.5B-12B
Mellum 2 Thinking is a Mixture-of-Experts reasoning model (64 experts, 8 activated per token, 131,072-token context) that emits its chain of thought. (Q4_K_M)
1,106 Pulls 1 Tag Updated 1 month ago
-
nanbeige4.1-python-deepthink
Fine-tuned version of Nanbeige 4.1 3B specialized for Python code generation with direct, focused output.
3b860 Pulls 3 Tags Updated 8 months ago
-
codellama-python-13b-q6
Specialized CodeLlama variant fine-tuned specifically for Python code generation. 13B params, Q6_K quant (very high quality, minimal loss).
698 Pulls 1 Tag Updated 8 months ago
-
gemma-4-E4B-it-textonly-sysprmpt-Q4_K_M
A structurally extracted, text-only iteration of Google's multimodal gemma-4-E4B-it model. Vision and audio encoders have been fully decoupled to minimize VRAM footprint for text-centric workloads. System Prompt to address lost abilities.
tools thinking665 Pulls 1 Tag Updated 6 months ago
-
Spark-X2.5-4B-Q6_K
Spark-X2.5-4B is a compact, general-purpose language model developed by SparkLLM. It outperforms similar models within the same parameter band.
624 Pulls 1 Tag Updated 1 month ago
-
arch-router
A specialized 1.5B parameter model for intelligent routing between multiple LLMs based on domain and action preferences.
1.5b452 Pulls 1 Tag Updated 9 months ago
-
gpt-oss-20B-Q5_K_M
12GB Version of GPT-OSS:20B via Unsloth for a slightly smaller footprint with minimal quality loss.
tools thinking300 Pulls 1 Tag Updated 1 month ago
-
squishy
A tiny 150M completion model trained from scratch for short story generation and small-model pipeline validation. It is best for generating small story prompts. (Think young child telling small stories. And cute at times.)
150m182 Pulls 1 Tag Updated 6 months ago
-
properly
Properly is the proofreader that doesn't steal your voice. Trained on Grammarly's CoEdit. LoRA was used with Gemma 1b-it.
106 Pulls 1 Tag Updated 6 months ago
-
Astrea-R8-Chat-9B
Astrea is a compact 9B-parameter model built for one thing: writing that feels written. It sounds human, it keeps the story straight, and it beats bigger models at both.
100 Pulls 1 Tag Updated 1 month ago
-
Groot-19B-A3B-261023
Groot is a family of open-weight reasoning models for complex real-world tasks. It scores higher than the larger Qwen3.6-35B-A3B and Gemma-4-26B-A4B, and it gives shorter answers than both. 19B total parameters and activates about 3B per token. Quant: 4_0
vision98 Pulls 1 Tag Updated 1 week ago
-
PentaCoder-9B
The foundational weights of Qwen/Qwen3.5-9B with selected specialization to maximize coding + agentic work at a small size using distillation/donar weights. ~7 GB loaded. Q4_K_M.
tools thinking97 Pulls 1 Tag Updated 3 weeks ago
-
Muse-Glimmer-30B-UD-Q3-K-XL
Unsloth UD Q3 Version of Muse Glimmer for a smaller footprint.
58 Pulls 1 Tag Updated 1 month ago
-
Deimos-A1
Deimos A1 is a concise chain-of-thought (CCoT) fine-tune of Qwen3.5-4B. It produces dense, stepwise <think> blocks averaging ~1/8 the tokens of the base model while improving accuracy on every reasoning benchmark measured.
50 Pulls 1 Tag Updated 6 months ago
-
HuatuoGPT-3-9B
HuatuoGPT-3-9B is a medical LLM built on Qwen3.5-9B with One-stage Policy Optimization (OnePO). OnePO adapts language models to medicine in a single reinforcement-learning stage, without preceding domain-specific supervised fine-tuning. Q4_K_M Version.
39 Pulls 1 Tag Updated 2 weeks ago
-
Moonlight-16B-A3B-Instruct
Moonlight, a 3B/16B-parameter Mixture-of-Expert (MoE) model trained with 5.7T tokens using Muon. Their "model improves the current Pareto frontier, achieving better performance with much fewer training FLOPs compared to prior models."
38 Pulls 1 Tag Updated 2 weeks ago
-
MINT-empathy-Qwen3-4B
MINT (Multi-turn Inter-tactic Novelty Training) model for empathic dialogue, fine-tuned from Qwen/Qwen3-4B.
37 Pulls 1 Tag Updated 6 months ago
-
muse-glimmer-30B-Q3-IQ3M
Q3 quantized variant of Muse Glimmer 30B for a slightly smaller footprint.
35 Pulls 1 Tag Updated 1 month ago
-
theprint-10B-MoE-A3B-0126-Q6_K
Llama Mix of Specialists with a small footprint.
tools11 Pulls 1 Tag Updated 1 month ago
-
Supra2-100M-Instruct-F16
Supra2-100M Instruct is a 100M-parameter decoder-only language model pretrained from scratch by SupraLabs on 30B tokens of English web text. It uses the Qwen3 architecture with a custom 32,768-token tokenizer.
9 Pulls 1 Tag Updated 2 weeks ago
-
mira-12B
A 12B chat model that talks like a person, not like an assistant. Short when you're short, warm when you're down, straight to the point when you're busy. No bullet-point essays, no "Great question!", no em dashes. It thinks before it answers.
tools thinking9 Pulls 1 Tag Updated 2 weeks ago