---
license: apache-2.0
base_model: Qwen/Qwen3.5-35B-A3B-Base
library_name: transformers
pipeline_tag: text-generation
datasets:
- nvidia/Nemotron-Terminal-Corpus
- open-thoughts/OpenThoughts-Agent-SFT-100K
tags:
- qwen3.5
- qwen
- terminal-agent
- terminus-2
- sft
---

# Qwen3.5-35B-A3B terminal agent — step 5,271

This is the step-5,271 intermediate checkpoint of the **terminal agent** run.
It is a supervised fine-tune of `Qwen/Qwen3.5-35B-A3B-Base` for agentic text generation.
Terminal-agent SFT over Terminus-2-focused trajectories from two modern teacher lineages.

This repository contains BF16 Hugging Face safetensors exported from the finalized
Megatron distributed checkpoint. It is not quantized.

## Checkpoint identity

| Field | Value |
| --- | --- |
| Hugging Face repository | `eewer/qwen3.5-35b-a3b-base-terminal-agent-step5271` |
| Run name | `terminal` |
| Training iteration | 5,271 |
| Planned iterations / epoch | 5,271 |
| Position in planned epoch | 100.00% |
| Base model | `Qwen/Qwen3.5-35B-A3B-Base` |
| Source checkpoint | `/KRAFTON/WORKSPACE/wbl-workspace/posttraining-2606/areal_runs/qwen35_b300/sft/combined-8gpu-tp1-ep8-cp1-mbs1-full-router1-shared0-gbs32-lr2e5/checkpoints/iter_0005271` |
| Export format | BF16 Hugging Face safetensors |
| Context length used for SFT | 32,768 tokens |

## Training configuration

- Hardware: one node with 8 NVIDIA B300 GPUs.
- Framework: Megatron-Bridge / Megatron-Core in the NVIDIA NeMo 26.06 environment.
- Parallelism: TP=1, PP=1, CP=1, EP=8, DP=8.
- Micro batch size: 1 per data-parallel rank.
- Global batch size: 32.
- Activation recomputation: full, uniform recomputation.
- MoE router fusion: enabled; shared-expert overlap, gradient-reduce overlap, and
  parameter-gather overlap disabled.
- Precision: BF16 training/model tensors with the recipe's precision-aware optimizer
  state; exported weights are BF16.
- Learning-rate schedule: cosine decay from a peak of `2e-5` to `2e-6`.
- Warmup: 158 optimizer steps for every run.
- Checkpoint interval: 500 optimizer steps.
- Sequence packing: No; each trajectory remains an independent 32K sample.
- Objective: assistant-only causal language modeling. System, user, and tool messages
  are context rather than prediction targets.

## Training data

The materialized source mixture has 168,646 rows; 168,646
rows enter training after the 32K boundary check. The complete training representation
contains 2,912,501,244 effective/rendered tokens.

| Source dataset | Selected rows |
| --- | ---: |
| `nvidia/Nemotron-Terminal-Corpus` | 79,201 |
| `open-thoughts/OpenThoughts-Agent-SFT-100K` | 89,445 |

The dataset is a deterministic shuffled union. Trajectory boundaries are preserved (no sequence packing). System and user tokens are context-only. Valid assistant turns are supervised; known malformed Nemotron recovery turns carry a per-turn zero loss mask while the surrounding valid assistant turns remain trainable.

The tokenizer is from Qwen3.5, with a Qwen3.6 thinking-preserving chat template. The
template keeps reasoning content in assistant messages and preserves tool declarations
when the source supplies a validated schema.

## Intended use

This checkpoint is intended for research on terminal, software-engineering, tool-use,
and interactive agents. Use the included chat template and supply tool schemas expected
by the target harness. It can be served with Transformers-compatible runtimes that
support the Qwen3.5 MoE architecture.

This is an intermediate training checkpoint, not a polished instruction model. Compare
checkpoints using held-out loss and task-level agent evaluations before selecting one
for deployment.

## Limitations and safety

- No complete benchmark suite is claimed in this model card.
- Agent trajectories can produce destructive shell commands, modify files, call tools,
  or expose secrets. Run the model in an isolated environment with scoped credentials.
- Tool names and schemas vary across source harnesses. A caller must provide the schema
  appropriate for its own environment; do not assume a generated call is valid or safe.
- Dataset filtering and deduplication reduce, but cannot guarantee removal of, benchmark
  contamination, incorrect reasoning, insecure code, or teacher-model artifacts.
- The model can hallucinate successful tool execution and should not be trusted without
  checking actual environment observations.
- This checkpoint inherits the base model's limitations and license.

## Reproducibility notes

All three runs use the same B300 execution topology, optimizer schedule, 158-step warmup,
and checkpoint cadence. The `terminal_swe` and `terminal_swe_general` runs differ from
the `terminal` run primarily in their source mixture and use of sequence packing. The
mixed datasets were validated for unique conversation hashes, deterministic shuffle
order, non-empty reasoning in every assistant turn, maximum rendered length, and
tool-call/schema consistency.