atrost/climbmix-matformer-353m-1p2b-h100-m

Extracted M MatFormer submodel from atrost/climbmix-matformer-353m-1p2b-h100.

Details

  • Architecture: ExtractedMatFormerForCausalLM
  • MatFormer size: m
  • Parameters: 190,741,760
  • Source checkpoint: atrost/climbmix-matformer-353m-1p2b-h100
  • Source checkpoint revision: 09960b1419b0340d8e918107e3915c0dccd852de
  • Source training actual tokens: 1,201,668,096
  • Slice spec: {'num_attention_heads': 4, 'num_key_value_heads': 2, 'intermediate_size': 1024}

Loading

from transformers import AutoModelForCausalLM, AutoTokenizer

repo_id = "atrost/climbmix-matformer-353m-1p2b-h100-m"
tokenizer = AutoTokenizer.from_pretrained(repo_id)
model = AutoModelForCausalLM.from_pretrained(
    repo_id,
    trust_remote_code=True,
    torch_dtype="auto",
)

trust_remote_code=True is required for the custom MatFormer extracted architecture.

Downloads last month
7
Safetensors
Model size
0.2B params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Dataset used to train atrost/climbmix-matformer-353m-1p2b-h100-m