This is Arm B of a size-matched ablation studying whether the gain from lowering the trajectory-inclusion threshold comes from data quality or data quantity. Arm B holds the sample count fixed at 2,040 (equal to the all-pass baseline, Arm A) but draws uniformly from the full trajectory pool, so only composition varies:

	trajectories	composition
Arm A (baseline)	2,040	100% fully-passing
Arm B (this model)	2,040	1,255 fully-passing + 785 imperfect (38%)

Training data

Trajectories are multi-turn bash-agent rollouts. Tool calls are emitted as plain text inside the assistant turn using a <tool_call>{"name": "bash", ...}</tool_call> protocol defined in the system prompt (not native function calling), and tool output returns in a tool role. Loss is computed on assistant turns only.

Training procedure

Full-parameter SFT via LLaMA-Factory, DeepSpeed ZeRO-3, BF16, on 8x B200.

hyperparameter	value
learning rate	5e-6
lr scheduler	cosine, warmup ratio 0.1
weight decay	0.0
optimizer	AdamW
epochs	5
per-device batch size	1
gradient accumulation	8
effective batch size	64
max sequence length	20,000
precision	bf16
attention	sdpa

Results

Final train loss 0.2652 (from 0.9019 at step 5); mean train loss 0.3865 over 160 steps / 5 epochs. Runtime 8,812s.

Note: 5.3% of samples exceed the 20,000-token cutoff and are truncated.

Framework versions

Transformers 5.8.0
PyTorch 2.11.0+cu130
DeepSpeed 0.18.9
Datasets 4.0.0
Tokenizers (via transformers)

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "March07/Qwen2.5-Coder-32B-terminal-agent-sft-mix2040"
tok = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, dtype="bfloat16", device_map="auto")

messages = [
    {"role": "system", "content": "You are an expert technical assistant with access to bash tools."},
    {"role": "user", "content": "List the files in /workspace."},
]
inputs = tok.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
out = model.generate(inputs, max_new_tokens=512)
print(tok.decode(out[0][inputs.shape[-1]:], skip_special_tokens=True))