Downloads · 30 days
0
Ventril
Skybox-Studios/Ventril
Ventril is a machine learning model from Skybox-Studios. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Access
Public
Updated Jul 5, 2026
Repo size
152 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors141 MB · 92%
At a glance
- Access
- Public
- Created
- Jul 4, 2026
- Updated
- Jul 5, 2026
- SHA
- f2c10779
Model card
From the Hugging Face model README
Ventril Specifications | Specification | Details | | :--- | :--- || | Total Parameters | 529 Million |
| Trainable (LoRA) Parameters | ~15 Million (Rank 64) |
| Context Window | 32,768 tokens | | Primary Task | Logical Reasoning & Step-by-Step Problem Solving | General Chatbot |
| Training Precision | bfloat16 |
| Target Modules | q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj |
| LoRA Alpha | 128 |
| Learning Rate | 2e-4 | | Datasets | GSM8K (Reasoning), OpenAssistant (Instruction following) |
| Special Tokens | <think>, </think>, Final Answer: |
Training Hyperparameters Batch Size: 4 (per device) Gradient Accumulation: 4 Total Training Steps: 1300+ (Combined aross training sessions) Optimizer: AdamW Weight Decay: 0.01 Most Recent Train Time and Checkpoint Count: Trained for 2H Checkpoint: 800
- Created
- Jul 4, 2026
- Updated
- Jul 5, 2026