Counter-SFT-01

Full-parameter BF16 supervised fine-tuning of Qwen3.5-0.8B on the Counter-SFT-01 dataset.

Results

Split Task success
Balanced train sample 65.6%
Validation 56.7%
Test 60.0%
Challenge template D 36.7%

Parse, format, allowed-action, budget, and final-field validity were 100% on every evaluated split. Arithmetic remained the main failure category.

Generation uses <|im_end|> token ID 248046 as EOS and PAD.

Downloads last month
18
Safetensors
Model size
0.9B params
Tensor type
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for ishagarg1103/counter-sft-01-model

Finetuned
(853)
this model

Dataset used to train ishagarg1103/counter-sft-01-model