luigi86/magnum-v2-72b_mlx-4bpw

MLX Format and Quantizations for Magnum v2 72b

Quantized to 4 bpw precision and tested using the mlx_lm utility on a 64GiB URAM M1 Max.

See original model for further details.

Larger, 8bpw quants available at mlx-community.

Original Model card

This is the seventh (Lucky!) in a series of models designed to replicate the prose quality of the Claude 3 models, specifically Sonnet and Opus. This model is fine-tuned on top of Qwen-2 72B Instruct.

Prompting

Model has been Instruct tuned with the ChatML formatting. A typical input would look like this:

"""<|im_start|>user
Hi there!<|im_end|>
<|im_start|>assistant
Nice to meet you!<|im_end|>
<|im_start|>user
Can I ask a question?<|im_end|>
<|im_start|>assistant
"""

Credits

• anthracite-org/Stheno-Data-Filtered
• anthracite-org/kalo-opus-instruct-22k-no-refusal
• anthracite-org/nopm_claude_writing_fixed

This model has been a team effort, and the credits goes to all members of Anthracite.

Training

The training was done for 2 epochs. We used 8x AMD Instinct™ MI300X Accelerators for the full-parameter fine-tuning of the model.

We also trained with a weight decay of 0.01 to help further stabilize the loss trajectory and mitigate catastrophic forgetting, and utilize a peak learning rate of 4e-6 to prevent the 2nd epoch loss from dropping too significantly (as it is a strong indicator of overfitting).

Sample Packing was done for 16k tokens rather than the 8k tokens used in our previous runs.

Safety

...

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	41.15
IFEval (0-Shot)	75.60
BBH (3-Shot)	57.85
MATH Lvl 5 (4-Shot)	31.65
GPQA (0-shot)	18.12
MuSR (0-shot)	14.18
MMLU-PRO (5-shot)	49.51

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	41.15
IFEval (0-Shot)	75.60
BBH (3-Shot)	57.85
MATH Lvl 5 (4-Shot)	31.65
GPQA (0-shot)	18.12
MuSR (0-shot)	14.18
MMLU-PRO (5-shot)	49.51