TIGER-Lab / FIM-8B

FIM-8B Inference on SWE-Bench Verified

This guide describes how to run the FIM-8B checkpoint on SWE-Bench Verified (and Lite). Unlike FIM-7B and FIM-14B (R2E-Gym scaffold), this model is post-trained on SWE-Lego trajectories and is evaluated with the SWE-Lego setup: OpenHands CodeActAgent for inference and the official SWE-bench harness for scoring.

Model

Local path: models/FIM-8B/ (checkpoints are gitignored; do not commit them).

• Base model: Qwen/Qwen3-8B
• FIM mid-training: FIM v2/v3 data
• Post-training: SFT on SWE-Lego trajectories

1. Serve the model with vLLM

CUDA_VISIBLE_DEVICES=0 \
python -m vllm.entrypoints.openai.api_server \
  --model models/FIM-8B \
  --served-model-name FIM-8B \
  --host 127.0.0.1 \
  --port 8400 \
  --tensor-parallel-size 1 \
  --max-model-len 163840 \
  --max-num-seqs 16 \
  --gpu-memory-utilization 0.9 \
  > vllm_fim8b.log 2>&1 &

The checkpoint ships max_position_embeddings: 163840 and its own chat template, so no rope or template overrides are needed.

Wait until the server is up (model load takes ~1 minute):

curl -s http://127.0.0.1:8400/v1/models

2. Run the agent on SWE-Bench Verified

Inference uses OpenHands 0.53.0. Define the LLM in config.toml :

[llm.eval_fim]
model = "openai/FIM-8B"
base_url = "http://127.0.0.1:8400/v1"
api_key = "EMPTY"
temperature = 0.0
max_input_tokens = 147456
max_output_tokens = 16384
native_tool_calling = false

From the OpenHands checkout:

env USE_HINT_TEXT=false \
    INSTRUCTION_TEMPLATE_NAME=swe_default.j2 \
    ENABLE_PLAN_MODE=false \
    ADD_IN_CONTEXT_LEARNING_EXAMPLE=false \
poetry run python evaluation/benchmarks/swe_bench/run_infer.py \
  --config-file config.toml \
  --agent-cls CodeActAgent \
  --llm-config llm.eval_fim \
  --max-iterations 100 \
  --eval-num-workers 1 \
  --eval-output-dir ./eval_out \
  --dataset princeton-nlp/SWE-bench_Verified \
  --split test \
  --mode swe

For SWE-Bench Lite, use --dataset princeton-nlp/SWE-bench_Lite .

3. Score with the SWE-bench harness

Convert the OpenHands output.jsonl to a predictions file with evaluation/benchmarks/swe_bench/scripts/eval/convert_oh_output_to_swe_json.py , then evaluate it with the official SWE-bench harness ( python -m swebench.harness.run_evaluation ). The reported score is resolved_instances / total_instances .