turbolego/ qwen3.5-0.8b-base-q4_k_m:latest

14 Downloads Updated 2 days ago

ollama run turbolego/qwen3.5-0.8b-base-q4_k_m

curl http://localhost:11434/api/chat \
  -d '{
    "model": "turbolego/qwen3.5-0.8b-base-q4_k_m",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='turbolego/qwen3.5-0.8b-base-q4_k_m',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'turbolego/qwen3.5-0.8b-base-q4_k_m',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 2 days ago

2 days ago

82356b9b61e8 · 533MB

model

archqwen35

·

parameters752M

·

quantizationQ4_K_M

533MB

template

{{ .Prompt }}

13B

Readme

Qwen3.5 0.8B Base — Q4_K_M

This is the unadapted Qwen3.5-0.8B comparison baseline for the AI-WCAG-Gauntlet benchmark. It is intended to be evaluated against the paired LoRA model to measure the effect of the LoRA changes on benchmark performance.

This model was imported from the Unsloth Q4_K_M GGUF release linked below. It is not the larger hybrid/speculative-decoding qwen3.5:0.8b Ollama library model.

Run

ollama run turbolego/qwen3.5-0.8b-base-q4_k_m

Comparison

For an A/B comparison, use the same benchmark instructions, tasks, tools, runtime settings, and evaluation process for this model and the LoRA variant. Keep the model as the only intentional variable, then compare the benchmark results, including the Agent Performance Score, validation and accessibility findings, HTML5 tag coverage, and iterations required.

The paired LoRA model is turbolego/qwen3.5-0.8b-wcag-gauntlet-lora-q4_k_m. Both models use Q4_K_M quantization and the same Ollama prompt template.

Reference

# Qwen3.5 0.8B Base — Q4_K_M

This is the unadapted Qwen3.5-0.8B comparison baseline for the [AI-WCAG-Gauntlet benchmark](https://github.com/turbolego/AI-WCAG-Gauntlet). It is intended to be evaluated against the paired LoRA model to measure the effect of the LoRA changes on benchmark performance.

This model was imported from the Unsloth Q4_K_M GGUF release linked below. It is not the larger hybrid/speculative-decoding `qwen3.5:0.8b` Ollama library model.

## Run

```bash
ollama run turbolego/qwen3.5-0.8b-base-q4_k_m
```

## Comparison

For an A/B comparison, use the same benchmark instructions, tasks, tools, runtime settings, and evaluation process for this model and the LoRA variant. Keep the model as the only intentional variable, then compare the benchmark results, including the Agent Performance Score, validation and accessibility findings, HTML5 tag coverage, and iterations required.

The paired LoRA model is [`turbolego/qwen3.5-0.8b-wcag-gauntlet-lora-q4_k_m`](https://ollama.com/turbolego/qwen3.5-0.8b-wcag-gauntlet-lora-q4_k_m). Both models use Q4_K_M quantization and the same Ollama prompt template.

## Reference

- [Unsloth Qwen3.5-0.8B GGUF](https://huggingface.co/unsloth/Qwen3.5-0.8B-GGUF)
- [AI-WCAG-Gauntlet benchmark](https://github.com/turbolego/AI-WCAG-Gauntlet)

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)