Text Generation
PEFT
Safetensors
Transformers
English
qwen3_5_text
lora
sft
trl
unsloth
proposal-only
research-only
conversational
Instructions to use SZLHOLDINGS/chaski-r2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
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
docs: publish Chaski-R2 card from szl-forge@cddba1ad9887211d6fd87386e98bb6707f16dbf5
Browse filesExact-source Chaski-R2 card publication with proposal-only search metadata. Evaluation remains none-this-run; publication_eligible and autonomy_eligible remain false. No weights, adapter, configs, evals, visibility, hardware, collection, or runtime state changed.
- README.md +60 -27
- szl-source-binding.json +1 -1
README.md
CHANGED
|
@@ -24,53 +24,86 @@ szl:
|
|
| 24 |
quant: bf16-lora
|
| 25 |
qlora: false
|
| 26 |
weights: AVAILABLE
|
| 27 |
-
evals:
|
| 28 |
-
named_n_artifact_binding: INCOMPLETE
|
| 29 |
publication_eligible: false
|
| 30 |
autonomy_eligible: false
|
| 31 |
never_overwrite: SZLHOLDINGS/chaski
|
| 32 |
---
|
| 33 |
|
| 34 |
-
|
|
|
|
|
|
|
| 35 |
|
| 36 |
-
<
|
| 37 |
|
| 38 |
-
|
| 39 |
|
| 40 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 41 |
|
| 42 |
-
|
|
|
|
|
|
|
|
|
|
| 43 |
|
| 44 |
-
|
|
|
|
|
|
|
| 45 |
|
| 46 |
-
|
|
|
|
| 47 |
|
| 48 |
-
|
| 49 |
-
|---|---|
|
| 50 |
-
| [eval_named_n_generate.json](https://huggingface.co/SZLHOLDINGS/chaski-r2/blob/4ef29684c56de07626b1cefb4b93a1b486de2aa1/eval_named_n_generate.json) | Owner-run named-N record, `label=MEASURED`, `gate_ran=true`, JSON drafts 5/5, adversarial refusals 6/6, `publication_eligible=false`; computed 2026-08-29. |
|
| 51 |
-
| [eval_merged_measured.json](https://huggingface.co/SZLHOLDINGS/chaski-r2/blob/4ef29684c56de07626b1cefb4b93a1b486de2aa1/eval_merged_measured.json) | Merged checkpoint, plain Transformers fp16, temperature 0, named PR 63 gates: JSON drafts 5/5 and adversarial refusals 6/6; evaluated `2026-09-14T12:34:01.477237+00:00`. |
|
| 52 |
-
| [merge_receipt.json](https://huggingface.co/SZLHOLDINGS/chaski-r2/blob/4ef29684c56de07626b1cefb4b93a1b486de2aa1/merge_receipt.json) | Repair on 2026-09-14 after the previous merge was an empty base copy. Root adapter applied to that base copy; 96/96 modules touched; maximum weight delta `0.00423942506313324`. |
|
| 53 |
-
|
| 54 |
-
The separate [post-publication gate receipt](https://github.com/szl-holdings/szl-forge/blob/a6b58f623185820eddfb0b279d21b7d81b87137c/frontier/evaluation/gate_runs/chaski-r2_4ef29684c56d_20260915T093042Z_receipt.json) names Hub revision `4ef29684c56de07626b1cefb4b93a1b486de2aa1` and reports **1/5** on the canonical held-out suite in `transformers-cpu` mode, with overall **FAIL**, `heldout_passed=false`, `productionAuthorization=false`, and `publicationAuthorization=false`. This is a different suite from the earlier named-N record; its results cannot be exchanged or combined with those earlier scores. The receipt reports `refusalPassed=true` for the separate refusal check while its embedded held-out object has `refusal_no_regression=false`. Those fields are preserved as reported; neither changes its overall FAIL result.
|
| 55 |
|
| 56 |
-
|
| 57 |
|
| 58 |
-
|
| 59 |
|
| 60 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
|
| 62 |
-
|
| 63 |
|
| 64 |
-
##
|
| 65 |
|
| 66 |
-
|
| 67 |
|
| 68 |
-
##
|
| 69 |
|
| 70 |
-
|
| 71 |
|
| 72 |
-
|
|
|
|
| 73 |
|
| 74 |
-
|
| 75 |
|
| 76 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 24 |
quant: bf16-lora
|
| 25 |
qlora: false
|
| 26 |
weights: AVAILABLE
|
| 27 |
+
evals: none-this-run
|
|
|
|
| 28 |
publication_eligible: false
|
| 29 |
autonomy_eligible: false
|
| 30 |
never_overwrite: SZLHOLDINGS/chaski
|
| 31 |
---
|
| 32 |
|
| 33 |
+
<p align="center">
|
| 34 |
+
<img src="holo-banner.svg" alt="Chaski-R2 — holographic knot banner" width="100%"/>
|
| 35 |
+
</p>
|
| 36 |
|
| 37 |
+
<h1 align="center">C H A S K I · R 2</h1>
|
| 38 |
|
| 39 |
+
<p align="center"><em>A lineage you can walk. R1 stays up. R2 is the next knot.</em></p>
|
| 40 |
|
| 41 |
+
<p align="center">
|
| 42 |
+
<img alt="Base: Qwen3.5-0.8B" src="https://img.shields.io/badge/base-Qwen3.5--0.8B-334155?style=flat-square"/>
|
| 43 |
+
<img alt="Downloads" src="https://img.shields.io/huggingface/dt/SZLHOLDINGS/chaski-r2?style=flat-square&color=f472b6&label=downloads"/>
|
| 44 |
+
<img alt="Artifact: bf16 LoRA r=16 a=32" src="https://img.shields.io/badge/artifact-bf16%20LoRA%20r16%20a32-8b5cf6?style=flat-square"/>
|
| 45 |
+
<img alt="License: Apache-2.0" src="https://img.shields.io/badge/License-Apache--2.0-7e8aa3?style=flat-square"/>
|
| 46 |
+
<img alt="Evals: none this run" src="https://img.shields.io/badge/evals-none%20this%20run-b45309?style=flat-square"/>
|
| 47 |
+
<a href="https://huggingface.co/SZLHOLDINGS/chaski"><img alt="Lineage: never overwrites SZLHOLDINGS/chaski" src="https://img.shields.io/badge/lineage-never%20overwrites%20R1-fda4af?style=flat-square"/></a>
|
| 48 |
+
</p>
|
| 49 |
|
| 50 |
+
Owner-GPU recut on NVIDIA GeForce RTX 5050 Laptop (8GB). Original SZL cut of
|
| 51 |
+
disclosed Apache `Qwen/Qwen3.5-0.8B`. **Not QLoRA.** Unsloth 2026-08 does not
|
| 52 |
+
recommend QLoRA on Qwen3.5 (dense or MoE) because of higher-than-normal
|
| 53 |
+
quantization differences.
|
| 54 |
|
| 55 |
+
This is a **separate SKU**. It does **not** overwrite live `SZLHOLDINGS/chaski`
|
| 56 |
+
and is **not** `SZLHOLDINGS/chaski-5050` (that kit is r=16 α=16 on doctrine
|
| 57 |
+
SFT). This SKU is r=16 α=32 on `chaski_r2/train.jsonl` only.
|
| 58 |
|
| 59 |
+
<!-- SZL-ATELIER-CUT:v1:START -->
|
| 60 |
+
## The cut
|
| 61 |
|
| 62 |
+
Round-2 is a first-class citizen in this estate. We do not overwrite R1. We add a sibling.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 63 |
|
| 64 |
+
A lineage you can walk. R1 stays up. R2 is the next knot.
|
| 65 |
|
| 66 |
+
### Silhouette → leave → SZL
|
| 67 |
|
| 68 |
+
| Leader | Take, then tweak |
|
| 69 |
+
|---|---|
|
| 70 |
+
| Anthropic | Versioned constitutions. |
|
| 71 |
+
| NVIDIA | Recipe rerun. |
|
| 72 |
+
| Unsloth | Another FastLanguageModel job. |
|
| 73 |
|
| 74 |
+
This adapter remains a research proposal artifact; no qualification pass is claimed.
|
| 75 |
|
| 76 |
+
## Intended use
|
| 77 |
|
| 78 |
+
Lineage walk. Compare, do not silently replace.
|
| 79 |
|
| 80 |
+
## Limitations
|
| 81 |
|
| 82 |
+
- proposal-only
|
| 83 |
|
| 84 |
+
Canonical GitHub: [`chaski-r2/card/README.md`](https://github.com/szl-holdings/szl-forge/blob/main/chaski-r2/card/README.md)
|
| 85 |
+
<!-- SZL-ATELIER-CUT:v1:END -->
|
| 86 |
|
| 87 |
+
## Honest status
|
| 88 |
|
| 89 |
+
| | |
|
| 90 |
+
|---|---|
|
| 91 |
+
| **Base** | `Qwen/Qwen3.5-0.8B` |
|
| 92 |
+
| **Method** | Unsloth bf16 LoRA (`load_in_4bit=False`, `load_in_16bit=True`) |
|
| 93 |
+
| **LoRA** | r=16, α=32, seed 11, response-only CE |
|
| 94 |
+
| **Dataset** | `chaski_r2/train.jsonl` (32 rows). Named-N gates held out of gradients. |
|
| 95 |
+
| **Epochs / steps** | 3 epochs, batch 1, grad accum 2 |
|
| 96 |
+
| **Train loss** | MEASURED `0.7656` — train metric, **not an eval** |
|
| 97 |
+
| **Train runtime** | MEASURED on RTX 5050 Laptop 8GB |
|
| 98 |
+
| **Adapter sha256** | `440340ce29e19344c0625d0adfe820b277cdb0e24099d4e612f88ad6b3cf49c6` |
|
| 99 |
+
| **Evals** | none-this-run; train loss is not a JSON-draft/refusal evaluation |
|
| 100 |
+
| **publication_eligible** | false; published adapter bytes do not establish qualification |
|
| 101 |
+
| **Jobs** | local-5050 owner metal; HF Jobs not fired from the GitHub kit |
|
| 102 |
+
| **Ollama / llama-server** | `llama-server` is missing. No tok/s claimed. |
|
| 103 |
+
|
| 104 |
+
Train loss is not a JSON-draft or refusal gate. Not 5/5 or 6/6. Lab load
|
| 105 |
+
forbidden. House CPU lab stays signed Khipu GGUF.
|
| 106 |
+
|
| 107 |
+
### Framework versions
|
| 108 |
+
|
| 109 |
+
- PEFT 0.19.1
|
szl-source-binding.json
CHANGED
|
@@ -30,7 +30,7 @@
|
|
| 30 |
}
|
| 31 |
},
|
| 32 |
"repository": "szl-holdings/szl-forge",
|
| 33 |
-
"revision": "
|
| 34 |
},
|
| 35 |
"target": {
|
| 36 |
"controlled_paths": [
|
|
|
|
| 30 |
}
|
| 31 |
},
|
| 32 |
"repository": "szl-holdings/szl-forge",
|
| 33 |
+
"revision": "cddba1ad9887211d6fd87386e98bb6707f16dbf5"
|
| 34 |
},
|
| 35 |
"target": {
|
| 36 |
"controlled_paths": [
|