---
base_model: Qwen/Qwen3-8B
library_name: peft
license: apache-2.0
tags:
- lora
- qlora
- qwen3
- codeact
- agents
- tool-use
- knapsack
- pythonformer
datasets:
- pythonformer/agents-learn-runtime-train
pipeline_tag: text-generation
---

# Qwen3-8B CodeAct Stateless LoRA — Opaque Knapsack (seed 777)

QLoRA (4-bit, r=64, alpha=128) fine-tune of [`Qwen/Qwen3-8B`](https://huggingface.co/Qwen/Qwen3-8B) on CodeAct
agent traces generated under a **stateless** Python interpreter runtime, trained to solve the
**Opaque Knapsack** task (partially-observable knapsack requiring iterative state revision via
a budgeted `inspect()` tool).

This is one run of a 2-seed variance sweep (`seed=777`, plus a sibling `seed=1337` run)
reproducing the original `seed=3407` result from the paper
**"Agents Learn Their Runtime: Interpreter Persistence as Training-Time Semantics"** (Victor May, Aaditya Salgarkar, Yishan Wang, Diganta Misra, Huu Nguyen, arXiv:2603.01209). The paper's core finding: a model
trained on **persistent**-runtime traces learns to rely on interpreter state persisting across
turns; when forced into a **stateless** runtime at inference it suffers cascading
`NameError`/recovery loops (the "amnesia tax"). See the [GitHub repo](https://github.com/pythonformer/pythonformer-workshop) for the full
pipeline.

## Training data

- Source: [`pythonformer/agents-learn-runtime-train`](https://huggingface.co/datasets/pythonformer/agents-learn-runtime-train),
  `data/knapsack/stateless/traces.jsonl` — **1000** CodeAct agent traces generated under the
  stateless runtime, teacher model Gemini (see paper Table 1, 17.8-22.8% teacher exact-solve rate).
- Copied byte-identical (sha256-verified) into the training pipeline, zero transformation.
- **Not** a re-derived/reconstructed trace set — this repo's earlier attempts to regenerate an
  equivalent set locally (`prepare_paired_data.py`, 1109 pairs) were found to NOT match this
  canonical data and were discarded; see `HANDOFF_knapsack_seed_sweep.md` in the repo for the
  full story.

## Hyperparameters

| | |
|---|---|
| Base model | `Qwen/Qwen3-8B` |
| Quantization | 4-bit QLoRA (bitsandbytes NF4) |
| LoRA rank / alpha / dropout | 64 / 128 / 0.05 |
| LoRA target modules | q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj |
| Training context | 16384 tokens |
| Batch size | 1 (micro) x 16 (grad accum) = 16 effective |
| Epochs | 3 (~45 optimizer steps over 1000 examples) |
| Optimizer / LR / schedule | AdamW (torch) / 1e-4 / cosine, 3% warmup |
| Training seed | **777** |
| Thinking | disabled (`enable_thinking: false`, i.e. `/nothink`) |

Final training loss: **0.1347** (this run) — manifest and full Axolotl config recorded in
`manifest.json` / `axolotl_config.yaml` in this repo, also mirrored into this HF repo for
provenance.

## Benchmark results (this checkpoint: stateless-trained, seed 777)

n=300 tasks/cell, from
[`pythonformer/agents-learn-runtime-tasks`](https://huggingface.co/datasets/pythonformer/agents-learn-runtime-tasks)
(easy/medium knapsack splits). "matched" = eval runtime equals the training runtime; "cross-eval"
= eval runtime differs from training runtime (tests the amnesia-tax hypothesis). Metric is
**score** (`metrics.score.mean`, partial-credit normalized optimality = agent's basket value /
optimal basket value, 0-1) — used instead of the stricter exact-optimum-match rate
(`average_is_solved`), which is too sparse at n=300 for this task to compare cleanly. Three
training seeds shown side by side, no averaging — **seed 3407** is the initial run (source:
[`pythonformer/agents-learn-runtime-benchmarks`](https://huggingface.co/datasets/pythonformer/agents-learn-runtime-benchmarks),
gated dataset, request access on the dataset page, n=300/cell), **seed 1337** / **seed 777** are
this variance sweep.

| eval runtime | difficulty | seed 3407 (initial) | seed 1337 | seed 777 |
|---|---|---|---|---|
| persistent (cross-eval) | easy | 0.294 | 0.143 | 0.155 |
| persistent (cross-eval) | medium | 0.092 | 0.015 | 0.032 |
| stateless (matched) | easy | 0.614 | 0.137 | 0.117 |
| stateless (matched) | medium | 0.331 | 0.014 | 0.002 |

Seed 3407 numbers use the dataset's folder naming `{runtime}_{model}_knapsack_{difficulty}`
— confirmed by downloading a cell's `config.yaml` and reading its `model:`/`persistent_state:`
fields directly (the dataset's own README states the opposite order, `{model}_{runtime}`,
but that does not match this snapshot's actual folder structure).

### Base model reference (seed-independent, not this checkpoint)

| eval runtime | difficulty | seed 3407 (initial) | seed 1337 / seed 777 (base, seed-independent) |
|---|---|---|---|
| persistent | easy | 0.050 | 0.061 |
| persistent | medium | 0.030 | 0.030 |
| stateless | easy | 0.034 | 0.035 |
| stateless | medium | 0.030 | 0.017 |

## Related checkpoints

- [`TieuDaoChanNhan/qwen3-8b-persistent-knapsack-lora-seed1337`](https://huggingface.co/TieuDaoChanNhan/qwen3-8b-persistent-knapsack-lora-seed1337)
- [`TieuDaoChanNhan/qwen3-8b-persistent-knapsack-lora-seed777`](https://huggingface.co/TieuDaoChanNhan/qwen3-8b-persistent-knapsack-lora-seed777)
- [`TieuDaoChanNhan/qwen3-8b-stateless-knapsack-lora-seed1337`](https://huggingface.co/TieuDaoChanNhan/qwen3-8b-stateless-knapsack-lora-seed1337)
- [`TieuDaoChanNhan/qwen3-8b-stateless-knapsack-lora-seed777`](https://huggingface.co/TieuDaoChanNhan/qwen3-8b-stateless-knapsack-lora-seed777)

## Citation

```bibtex
@article{pythonformer260301209,
  title   = {Agents Learn Their Runtime: Interpreter Persistence as Training-Time Semantics},
  author  = {Victor May, Aaditya Salgarkar, Yishan Wang, Diganta Misra, Huu Nguyen},
  journal = {arXiv preprint arXiv:2603.01209},
  year    = {2026}
}
```