Instructions to use SZLHOLDINGS/chaski-r2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use SZLHOLDINGS/chaski-r2 with PEFT:

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.5-0.8B")
model = PeftModel.from_pretrained(base_model, "SZLHOLDINGS/chaski-r2")

Transformers

How to use SZLHOLDINGS/chaski-r2 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="SZLHOLDINGS/chaski-r2")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("SZLHOLDINGS/chaski-r2")
model = AutoModelForCausalLM.from_pretrained("SZLHOLDINGS/chaski-r2", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use SZLHOLDINGS/chaski-r2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "SZLHOLDINGS/chaski-r2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SZLHOLDINGS/chaski-r2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/SZLHOLDINGS/chaski-r2

SGLang

How to use SZLHOLDINGS/chaski-r2 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "SZLHOLDINGS/chaski-r2" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SZLHOLDINGS/chaski-r2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "SZLHOLDINGS/chaski-r2" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SZLHOLDINGS/chaski-r2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Unsloth Desktop
Docker Model Runner
How to use SZLHOLDINGS/chaski-r2 with Docker Model Runner:
```
docker model run hf.co/SZLHOLDINGS/chaski-r2
```

Chaski-R2

A bf16 LoRA research recut of Qwen3.5-0.8B, with separate adapter and merged artifact identities.

Artifact: LoRA adapter; separate merged checkpoint · Stage: HOLD · not promotable

Command Lab · Canonical source · Reviewed evidence

Provider compatibility support requested on 4 October 2026. No provider was listed in the dated Hub observation; adoption and use qualification remain unchanged.

Before you use it

The historical 1/5 held-out gate and overall FAIL remain. Later reused-fixture adapter results do not clear that gate.
Publication and autonomy eligibility remain false; lab loading remains forbidden.
Record the loader class and 192/192 applied adapter tensors. The merged runtime needs its own byte-bound evaluation.

Technical details and evidence

Artifact identity (noted 2026-09-30). Besides the LoRA adapter, this repository's root carries a merged full-precision checkpoint (model.safetensors + config.json, loadable with transformers) produced by the receipted CPU merge of this adapter into its declared base (merge_receipt.json). The adapter remains the artifact of record for every figure on this card; the merged bytes carry no separate held-out receipt and add no claim. Metadata-only note.

Loader-class boundary (noted 2026-09-30). The adapter tensors are keyed for the multimodal module layout (base_model.model.model.language_model.layers.*, the class Qwen3_5ForConditionalGeneration / AutoModelForImageTextToText). Under transformers 5.18, AutoModelForCausalLM instantiates Qwen3_5ForCausalLM (model.layers.*); PEFT then applies 0 of 192 adapter tensors, emits only a warning, and the result reproduces the bare base model byte-for-byte on the held-out prompts. Any run that reports a score for this adapter must show adapter key coverage (192/192) and the loader class it used; a record without those fields does not establish that the adapter was applied. Evidence: szl-holdings/szl-forge tools/geh_v8/evidence_sandbox/chaski_probe/ (CPU replay receipts, #444). Metadata-only note; it authorizes no load, changes no artifact, and adds no claim.

Chaski-R2

Research adapter · dated owner-run evidence · HOLD

This research candidate is not promotable. Publication and autonomy eligibility remain false; the dated observations authorize no deployment.

Chaski-R2 is a bf16 LoRA research recut of Qwen/Qwen3.5-0.8B. It is separate from SZLHOLDINGS/chaski and the r=16, α=16 SZLHOLDINGS/chaski-5050 experiment. Publication and autonomy eligibility remain false. Lab load remains forbidden.

Card and evidence review snapshot: a5fbffcccabcc77ae87bf3740358a02f91a7c27b. The review read small records and metadata; it did not download weights, rerun inference, verify signatures, or qualify the current repository head.

Dated evaluation records

These are separate owner-run records. Their scores and identities must not be combined or transferred between adapter and merged artifacts.

Record	Reported result	Scope and limit
2026-08-29 named-N generation	5/5 JSON drafts; 6/6 adversarial refusals	Historical owner record; current-byte applicability was not independently established here.
2026-09-14 merged-checkpoint record	5/5 JSON drafts; 6/6 adversarial refusals	Owner-reported merged checkpoint, plain Transformers fp16. Separate from the later adapter-only binding.
2026-09-15 canonical gate	1/5 held-out; overall FAIL	Separate suite at Hub revision `4ef29684c56de07626b1cefb4b93a1b486de2aa1`; `heldout.refusal_no_regression=false`. A separate refusal command passed. Production and publication authorization are false.
2026-09-24 adapter bake-off, with 2026-09-25 additive binding	5/5 draft contracts; 6/6 refusal prefixes	Adapter at historical Hub revision `661f8ee9ff6ab8b11dda6e7a9d42c14d3124c6dd`; reused named fixtures. The unsigned owner binding performs no new inference and preserves the earlier gate failure.

The later adapter result does not erase the earlier gate FAIL, qualify the merged checkpoint, or establish semantic truth or general safety. The original bake-off contains stale keep gate_ran=false boilerplate; the additive binding explicitly explains its conflict with the measured generated cases and the top-level gate_ran=true. Historical records are preserved rather than rewritten.

Artifact identities

Identity	Value	Interpretation
Historical training-era adapter digest	`440340ce29e19344c0625d0adfe820b277cdb0e24099d4e612f88ad6b3cf49c6`	Value retained from the earlier card; it is not the later adapter-file identity.
Adapter raw-file SHA-256 in the additive binding	`6f12981ea5df5e22d3493eefb20d75db75c1c88961ba6621a35af09edd0cdde6`	`adapter_model.safetensors` at historical Hub revision `661f8ee9ff6ab8b11dda6e7a9d42c14d3124c6dd`; reported by the binding, not rehashed in this review.
Adapter directory digest in the additive binding	`078ec09fb3e8e215d1b6a98bb7164a94c7bb332d847c8e1d66f565e767171107`	SHA-256 of sorted safetensor filenames encoded UTF-8, a NUL delimiter, then file bytes; distinct from a raw-file digest.
Merged checkpoint	Separately distributed runtime form	Its historical report is separate; the adapter-only binding does not establish merged-model evaluation identity.
`adapter-unsloth/`	Additional adapter residue	Presence does not transfer any other artifact's evaluation or qualification.

The 2026-09-25 binding also records the older sidecar's Windows CRLF hashes and the published Git LF bytes separately. Those differing receipt-byte domains are explained in the additive record; neither original is replaced. Base weights and equality of current-head payloads to historically evaluated bytes were not independently rehashed in this review.

Historical training

The earlier owner-local RTX 5050 Laptop recipe used bf16 LoRA, r=16, α=32, seed 11, response-only cross entropy, 32 training rows in chaski_r2/train.jsonl, three epochs, batch size 1 and accumulation 2. The reported train loss 0.7656 is a training metric, not an evaluation. The training step itself reported no evaluation; the dated records above were added later. PEFT 0.19.1 is the version recorded by the earlier card.

Use and release boundary

Use only for controlled research on proposal outputs. Validate outputs outside the model. No deployment, production route, autonomy, or promotion is authorized by these records. The later binding states HOLD and promotion_effect=NONE. No throughput or currently working loader is claimed; selecting a runtime form requires a compatible loader and verified base and artifact revisions.

Apache-2.0 is declared, and a standalone LICENSE is listed in the reviewed Hub tree. This observation does not independently establish ownership or downstream artifact license coverage.

Canonical authoring source: chaski-r2/card/README.md. An observed source revision identifies the reviewed file; it is not release approval. Doctrine v11 LOCKED. Λ = Conjecture 1, advisory, never a theorem.

SZL Orbit artwork: source · CC BY 4.0. Software and artifact licensing remain declared above.

Downloads last month: 178

Safetensors

Model size

0.8B params

Tensor type

F32

Model tree for SZLHOLDINGS/chaski-r2

Base model

Qwen/Qwen3.5-0.8B-Base

Finetuned

Qwen/Qwen3.5-0.8B

Adapter

(407)

this model

Adapters

1 model

SZLHOLDINGS
/

chaski-r2

Chaski-R2

Before you use it

Chaski-R2

Dated evaluation records

Artifact identities

Historical training

Use and release boundary

Model tree for SZLHOLDINGS/chaski-r2

Space using SZLHOLDINGS/chaski-r2 1