Views
No views yet
Qwen/Qwen3.5-4B / Qwen/Qwen3.5-4B-Baseqwen35).gguf)WithinUsAI/claude_mythos_distilled_25k - 25k samples optimized for stylized narrative and rich conversational style.11-47/claude_opus_4.8_max_thinking_5k_v2 - 5k samples targeting deep synthetic reasoning, structural text planning, and long-form consistency.11-47/claude_opus_mythos_5k - 5k samples specialized in deep narrative building and expressive prompt adherence.llama.cpp or CLI1# Start an OpenAI-compatible API server locally
2llama-server -hf WithinUsAI/Opus4.8-Enemey.Of.Thy.State-4B.gguf:Q4_K_M
3
4# Or run directly in your terminal terminal
5llama-cli -hf WithinUsAI/Opus4.8-Enemey.Of.Thy.State-4B.gguf:Q4_K_M
6llama-cpp-python1from llama_cpp import Llama
2
3llm = Llama.from_pretrained(
4 repo_id="WithinUsAI/Opus4.8-Enemey.Of.Thy.State-4B.gguf",
5 filename="Opus4.8-Enemey.Of.Thy.State-4B-Q4_K_M.gguf",
6)
7
8response = llm.create_chat_completion(
9 messages=[
10 {"role": "system", "content": "You are a creative narrative companion."},
11 {"role": "user", "content": "Write an opening scene for a political thriller."}
12 ]
13)
14print(response['choices'][0]['message']['content'])
15ollama run hf.co/WithinUsAI/Opus4.8-Enemey.Of.Thy.State-4B.gguf:Q4_K_M)
### Why this structure works perfectly for your model:
1. **The Metadata Block (YAML Frontmatter):** Hugging Face scans the code block at the absolute top (`---` to `---`). Adding your explicit training datasets like `WithinUsAI/claude_mythos_distilled_25k` guarantees your model will show up under those dataset pages automatically, driving more discoverability.
2. **Clear Application Focus:** Highlighting "Refusal Removal" and "Enhanced Prose" flags precisely what local LLM enthusiasts look for in a Qwen 3.5 4B fine-tune.