Downloads · 30 days
8
18% of all-time downloads
johanna09/qwen38-27b-cybertruck-rally-full
qwen38-27b-cybertruck-rally-full is a text generation model from johanna09. Use it when you need the model to write or continue text. It is set up for peft. The card lists the license as apache-2.0.
A LoRA adapter for Qwen3.8-27B trained to reliably generate a complete, playable Tesla Cybertruck monster-truck rally game as a single HTML file.
Downloads · 30 days
8
18% of all-time downloads
All-time downloads
45
Public
Repo size
658 MB
Likes
1
Public
Click a slice to open those files.
.safetensors638 MB · 97%
From the Hugging Face model README
A LoRA adapter for Qwen3.8-27B trained to reliably generate a complete, playable Tesla Cybertruck monster-truck rally game as a single HTML file.
The adapter is intentionally narrow. It was trained on 120 working game examples rather than on a general-purpose instruction dataset. The goal was to test how much a small, highly repetitive fine-tune can change a model's ability to produce a specific artifact, and what side effects that produces.
With the prompt:
Make a Cybertruck rally game.
the adapter typically generates roughly 1,400–1,600 lines of HTML (about 14k tokens) for a self-contained game.
The generated file includes:
The setting can also be varied through the prompt. For example, the training distribution includes beaches in the UAE, nighttime floodlit tracks, desert settings at dusk, and enclosed indoor arenas.
| Parameter | Value |
|---|---|
| Base model | unsloth/Qwen3.8-27B-unsloth-bnb-4bit |
| Training examples | 120 |
| Epochs | 3 |
| Optimizer steps | 90 |
| Effective batch size | 4 |
| LoRA rank / alpha | 32 / 64 |
| Trainable parameters | 159,383,552 (0.58% of 27,516,112,112) |
| Learning rate | 2e-4 |
| Maximum sequence length | 16,384 |
| Loss | Assistant responses only |
| System prompts | None |
| First-step loss | 0.3744 |
| Final-step loss | 0.00424 |
| Mean loss | 0.0419 |
| Training time | 1 h 41 m (66.9 s/it) |
| Peak VRAM | 34 GB |
Training was performed on 2026-08-26 with Unsloth on a single RTX PRO 6000 Blackwell (96 GB), using about $7 of rented GPU time.
The 120 examples were generated from a library of approximately 33 working Three.js components. The examples were varied across environment, arena shape, obstacle mix, truck count, finish conditions, camera rig, crowd style, and HUD.
Each example was tested in headless Chrome before being included in the training set:
All 120 examples passed these checks.
The examples are intentionally highly repetitive: they share roughly 90% of their code. This explains both the very low final training loss and the adapter's strong specialization.
Three settings are important when serving this adapter.
--max-num-seqsQwen3.8-27B uses linear-attention layers that require Mamba cache blocks. vLLM's default concurrency can exceed the available cache capacity.
For example:
ValueError: max_num_seqs (1024) exceeds available Mamba cache blocks (469)
Set --max-num-seqs explicitly.
The training data used Qwen's non-thinking format. At inference, use:
"chat_template_kwargs": {"enable_thinking": false}
Otherwise the model is being served with a different prompt format from the one used during training.
On Blackwell GPUs, set:
VLLM_USE_FLASHINFER_SAMPLER=0
Verified with vLLM 0.28.0:
VLLM_USE_FLASHINFER_SAMPLER=0 \
python -m vllm.entrypoints.openai.api_server \
--model unsloth/Qwen3.8-27B \
--served-model-name qwen3.8-27b \
--max-model-len 32768 \
--max-num-seqs 16 \
--enable-lora \
--max-lora-rank 32 \
--trust-remote-code \
--lora-modules cybertruck-rally-full=/path/to/this/adapter
Then send a request with no system message:
curl http://localhost:8000/v1/chat/completions \
-H "Content-Type: application/json" -d '{
"model": "cybertruck-rally-full",
"messages": [{"role": "user", "content": "Make a Cybertruck rally game."}],
"max_tokens": 24000,
"temperature": 0.8,
"chat_template_kwargs": {"enable_thinking": false}
}'
For a useful comparison, run the same request against the base model
qwen3.8-27b with the same prompt, temperature, and token budget.
The bf16 base model requires approximately 80 GB of VRAM. An H100 takes about 215 seconds for a cold start in the tested setup.
The fine-tune does more than teach the model the game. It also transfers some of the formatting and writing conventions that appear consistently in the training data.
For example:
| Pattern | Training data | Base model |
|---|---|---|
charset="utf-8" | 120/120 | charset="UTF-8" |
initial-scale=1 | 120/120 | initial-scale=1.0 |
* { margin: 0; padding: 0; box-sizing: border-box; } | 120/120 | Varies |
These patterns can appear even when the adapter is asked to generate an unrelated HTML artifact.
This is the main trade-off of the experiment: the adapter strongly reinforces the target artifact and also transfers some of its stylistic fingerprints.
This is a deliberately narrow post-training experiment: how far can a small LoRA fine-tune push a model toward a highly specific, impressive-looking artifact?
In this case, about $7 of GPU time and 90 optimizer steps were enough to make Qwen3.8-27B consistently generate a complete 3D rally game on demand.
The adapter is intended as a reproducible example of both the power and the limitations of narrow fine-tuning.