c4tdr0ut/grok-oss-Revenant-70B
Grok OSS Revenant 70B is a 70 billion parameter large language model from c4tdr0ut, distilled from Grok's voice mode conversations. It is fine-tuned using SFT and ORPO, offering enhanced reasoning, coherence, and instruction following. This model is specifically designed for users seeking maximum capability combined with an unfiltered, direct, and uncensored personality.
Loading preview...
Grok OSS Revenant 70B: Unfiltered Capability
Grok OSS Revenant 70B is the large-scale iteration in the Grok OSS Revenant series, developed by c4tdr0ut. This 70 billion parameter model was created by distilling raw, unfiltered conversations from Grok's voice mode, following the same training methodology as its 8B counterpart.
Key Capabilities
- Enhanced Reasoning: Offers significantly improved reasoning abilities compared to smaller models.
- High Coherence: Delivers more consistent and logical outputs.
- Strong Instruction Following: Excels at understanding and executing complex instructions.
- Unfiltered Personality: Retains a raw, unhinged, and uncensored conversational style, derived from its unique training data.
Training Process
The model underwent a two-stage training process:
- Stage 1 – Supervised Fine-Tuning (SFT): Utilized a high-quality multi-turn conversational dataset sourced from Grok voice mode.
- Stage 2 – ORPO: Further refined with a 1,000-sample preference dataset, incorporating filtered voice conversations and high-quality alignment data.
Training was conducted using QLoRA (4-bit) for approximately 2 hours on NVIDIA B200 hardware.
Good For
- Users requiring maximum capability in reasoning and instruction following.
- Applications where an unfiltered, direct, and uncensored personality is desired.
- Complex reasoning tasks that benefit from a large-scale model.
Limitations
- Can exhibit chaotic and vulgar language due to its training data.
- Not suitable for safe or professional use cases.