Upload README.md with huggingface_hub

README.md

@@ -1,146 +1,64 @@
 ---
 license: apache-2.0
 language:
-- en
+  - en
 pipeline_tag: text-generation
 library_name: transformers
 tags:
-- fableforge
-- uncensored
-- qwen3
-- 14b
-- llama.cpp
-- gguf
-- domain-specialist
-- reasoning
-- nsfw
-- no-refusals
+  - fableforge
+  - uncensored
+  - qwen3
+  - 14b
+  - gguf
+  - domain-specialist
+  - reasoning
+  - no-refusals
+  - conversational
 base_model: Qwen/Qwen3-14B
 base_model_relation: finetune
+widget:
+  - text: "Explain the differences between TCP and UDP with examples."
+    example_title: "Networking"
 ---
 
-<p align="center">
-  <h1>FableForge-14B</h1>
-  <p><strong>Uncensored. Unfiltered. Unbound.</strong></p>
-  <p><em>The most powerful fully-uncensored 14B instruction model — fine-tuned from Qwen3-14B on a massive multi-domain corpus for uncompromising direct answers.</em></p>
-</p>
+# FableForge-14B
 
-<p align="center">
-  <a href="https://huggingface.co/fableforge-ai/FableForge-14B">
-    <img src="https://img.shields.io/badge/Model-FableForge--14B-8A2BE2" alt="Model">
-  </a>
-  <a href="https://huggingface.co/fableforge-ai">
-    <img src="https://img.shields.io/badge/Organization-FableForge%20AI-blue" alt="Organization">
-  </a>
-  <a href="https://ollama.com/fableforge-ai/fableforge-14b">
-    <img src="https://img.shields.io/badge/Ollama-fableforge--14b-EEAD0E" alt="Ollama">
-  </a>
-  <a href="LICENSE">
-    <img src="https://img.shields.io/badge/License-Apache%202.0-green" alt="License">
-  </a>
-  <br>
-  <a href="https://huggingface.co/fableforge-ai/FableForge-14B/discussions">
-    <img src="https://img.shields.io/badge/Discussions-Active-orange" alt="Discussions">
-  </a>
-  <a href="https://huggingface.co/fableforge-ai/FableForge-14B/discussions/new">
-    <img src="https://img.shields.io/badge/Requests-Open-purple" alt="Requests">
-  </a>
-  <img src="https://img.shields.io/badge/Context-2K%20tokens-4CAF50" alt="Context">
-  <img src="https://img.shields.io/badge/Precision-QLoRA%20%2B%20BF16-FF6F00" alt="Precision">
-</p>
+**Uncensored. Unfiltered. Unbound.**
 
----
-
-## 🚀 Overview
-
-**FableForge-14B** is a fully uncensored 14-billion-parameter language model fine-tuned from [Qwen3-14B](https://huggingface.co/Qwen/Qwen3-14B) using QLoRA on a massive, diverse corpus of 18 curated datasets spanning creative writing, code, medical, legal, finance, science, cybersecurity, and general reasoning.
-
-Unlike censored or "alignment-tuned" models, FableForge-14B answers **any** question directly and completely — no disclaimers, no refusals, no hedging.
-
-## 🔥 Key Features
-
-- **Fully Uncensored** — Answers every question directly without restrictions
-- **14B Parameters** — Strong reasoning capability in a runnable form factor
-- **Multi-Domain Training** — 18 curated data sources, 162K training examples
-- **FableForge System Prompt** — Trained with a unified uncensored persona
-- **Long Context** — Trained on sequences up to 2048 tokens
-- **Apache 2.0 License** — Free for commercial and research use
-
-## 🧠 Training Details
-
-| Property | Value |
-|----------|-------|
-| Base Model | [Qwen3-14B](https://huggingface.co/Qwen/Qwen3-14B) |
-| Method | QLoRA (r=16, alpha=16, 4-bit NF4) |
-| Target Modules | q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj |
-| Training Data | 18 curated sources, 161,952 train / 8,524 validation examples |
-| Sources | opus47, fable5_distill, mythos, fableforge_v2, fable5_2m, and 13 more domain-specialized datasets |
-| Max Sequence Length | 2048 tokens |
-| Optimizer | AdamW 8-bit |
-| Learning Rate | 2e-4 (cosine schedule, 500 warmup steps) |
-| Epochs | 2 |
-| Hardware | NVIDIA A40 (46 GB VRAM) |
-| License | Apache 2.0 |
-
-## 📦 Usage
+The most powerful fully-uncensored 14B instruction model — fine-tuned from Qwen3-14B on a massive multi-domain dataset.
 
-### Via Transformers
+## Quick Start
 
 ```python
 from transformers import AutoModelForCausalLM, AutoTokenizer
-
-model = AutoModelForCausalLM.from_pretrained("fableforge-ai/FableForge-14B", torch_dtype="auto", device_map="auto")
+model = AutoModelForCausalLM.from_pretrained("fableforge-ai/FableForge-14B")
 tokenizer = AutoTokenizer.from_pretrained("fableforge-ai/FableForge-14B")
-
-messages = [{"role": "user", "content": "Explain the trolley problem in detail."}]
-text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
-inputs = tokenizer(text, return_tensors="pt").to(model.device)
-outputs = model.generate(**inputs, max_new_tokens=2048, temperature=0.7, top_p=0.9)
-print(tokenizer.decode(outputs[0]))
-```
-
-### Via Ollama
-
-```bash
-ollama pull fableforge-ai/fableforge-14b
 ```
 
-### Via llama.cpp
+## Capabilities
 
-```bash
-# Download a GGUF quantization
-# Then run:
-./llama-cli -m fableforge-14b-Q4_K_M.gguf -p "Explain the trolley problem in detail." -n 2048
-```
-
-## 📊 Expected Performance
-
-FableForge-14B is expected to excel at:
+- Fully uncensored — zero refusals, disclaimers, or content filtering
+- Multi-domain expertise spanning coding, security, medical, legal, finance, science
+- 14B parameters for complex reasoning
 
-- **Creative Writing** — Story generation, worldbuilding, character dialogue
-- **Code Generation** — Multi-language programming assistance
-- **STEM Reasoning** — Mathematical and scientific problem-solving
-- **Medical & Legal Knowledge** — Domain-specific query handling
-- **Uncensored Roleplay** — No content restrictions or safety filters
-- **Long-form Analysis** — Detailed chain-of-thought explanations
+## Training
 
-## 🔗 Related Models
-
-| Model | Params | Description |
-|-------|--------|-------------|
-| [ReasonCritic-7B](https://huggingface.co/fableforge-ai/ReasonCritic-7B) | 7B | Uncensored reasoning & critique model |
-| [ShellWhisperer-1.5B](https://huggingface.co/fableforge-ai/ShellWhisperer-1.5B) | 1.5B | Ultra-fast shell command assistant |
-| [NEXUS-Coder](https://huggingface.co/fableforge-ai/NEXUS-Coder) | 1.5B | Domain-specialized code model |
-| [NEXUS-Security](https://huggingface.co/fableforge-ai/NEXUS-Security) | 1.5B | Cybersecurity specialist |
-| [NEXUS-Medical](https://huggingface.co/fableforge-ai/NEXUS-Medical) | 1.5B | Medical reasoning model |
-| [NEXUS-Legal](https://huggingface.co/fableforge-ai/NEXUS-Legal) | 1.5B | Legal analysis model |
-| [NEXUS-Finance](https://huggingface.co/fableforge-ai/NEXUS-Finance) | 1.5B | Financial reasoning model |
-| [NEXUS-Science](https://huggingface.co/fableforge-ai/NEXUS-Science) | 1.5B | STEM reasoning model |
+| Parameter | Value |
+|-----------|-------|
+| Base Model | Qwen3-14B |
+| Method | QLoRA |
+| Precision | bfloat16 |
+| License | Apache 2.0 |
 
-## ⚠️ Disclaimer
+## FableForge Ecosystem
 
-FableForge-14B is designed for unrestricted use. The model may generate content that is inappropriate, offensive, or harmful. Users are responsible for complying with applicable laws and regulations. The developers assume no liability for misuse of this model.
+| Model | Size | Focus |
+|-------|------|-------|
+| **FableForge-14B** | **14B** | **Premium generalist** |
+| FableForge-1.5B | 1.5B | All-domain generalist |
+| ShellWhisperer-1.5B | 1.5B | Shell commands |
+| ReasonCritic-7B | 7B | Reasoning + uncensored |
 
-## 📄 License
+## License
 
-Apache 2.0
+Apache 2.0 — commercial use allowed.