Dataset · Hugging Face
armand0e/Fable-5-Chat
TheFusionCube Fable-5 Chat Conversion Source dataset: TheFusionCube/Fable-5-CoT-Traces Output file: train.jsonl Source rows: 468 Kept rows: 353 Dropped category == "decoy" rows: 115 Dropped blank prompt/response rows: 0 Each row has: { "prompt": "...", "messages": [ {"role": "user", "content": "..."}, {"role": "assistant", "content": "..."} ], "tools": [], "metadata": {"trace_type": "chat", "category": "..."} }
| Author / organization | armand0e |
|---|---|
| Downloads | 226 (as reported by Hugging Face) |
| Likes | 26 |
| Created | 22 June 2026 |
| Last updated at source | 22 June 2026 |
Popularity, last 2 days
226 now. Counts as reported by the source, snapshotted daily by Silytics.
Connected in the knowledge graph
- caslca/Qwen3.8-27B-Fable-Distill-OptiQ-4.5bpw-mixedModel · used to train
- mradermacher/Quartz-R1-8B-Genesis-i1-GGUFModel · used to train
- mradermacher/Quartz-R1-8B-Genesis-GGUFModel · used to train
- TeichAI/Qwen3.8-27B-Fable-Distill-GGUFModel · used to train
- caslca/Qwen3.8-27B-Fable-Distill-mlx-uniform-4bitModel · used to train
- mradermacher/Qwen3.6-35B-A3B-Fable-5-Distill-i1-GGUFModel · used to train
- TeichAI/Qwen3.8-27B-Fable-DistillModel · used to train
- McG-221/Qwen3.8-27B-Fable-Distill-mlx-8BitModel · used to train
- vwdubb/Qwen3.8-27B-Fable-Distill-FP8Model · used to train
- vwdubb/Qwen3.8-27B-Fable-Distill-NVFP4Model · used to train
- mradermacher/Qwen3.8-27B-Fable-Distill-i1-GGUFModel · used to train
- mradermacher/Qwen3.6-35B-A3B-Fable-5-Distill-GGUFModel · used to train
- mradermacher/Qwen3.8-27B-Fable-Distill-GGUFModel · used to train
More from armand0e
Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.