Instructions to use my-ai-stack/Stack-2-9-finetuned with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use my-ai-stack/Stack-2-9-finetuned with Transformers:

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

pipe = pipeline("text-generation", model="my-ai-stack/Stack-2-9-finetuned")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("my-ai-stack/Stack-2-9-finetuned")
model = AutoModelForCausalLM.from_pretrained("my-ai-stack/Stack-2-9-finetuned")
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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use my-ai-stack/Stack-2-9-finetuned with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "my-ai-stack/Stack-2-9-finetuned"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "my-ai-stack/Stack-2-9-finetuned",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/my-ai-stack/Stack-2-9-finetuned

SGLang

How to use my-ai-stack/Stack-2-9-finetuned 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 "my-ai-stack/Stack-2-9-finetuned" \
    --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": "my-ai-stack/Stack-2-9-finetuned",
		"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 "my-ai-stack/Stack-2-9-finetuned" \
        --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": "my-ai-stack/Stack-2-9-finetuned",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use my-ai-stack/Stack-2-9-finetuned with Docker Model Runner:
```
docker model run hf.co/my-ai-stack/Stack-2-9-finetuned
```

Stack-2-9-finetuned / stack /docs /reference /BENCHMARKS.md

walidsobhie-code

refactor: Squeeze folders further - cleaner structure

65888d5 about 2 months ago

preview code

raw

history blame

4.17 kB

	# Benchmark Results - Stack 2.9

	> Note: These benchmarks are currently in progress. Results will be published after training is complete.

	## Benchmark Overview

	Stack 2.9 will be evaluated on a comprehensive suite of benchmarks to measure coding capabilities, tool use proficiency, and overall model performance. The evaluation framework includes both standard coding benchmarks and custom tool-use scenarios.

	## Planned Benchmarks

	### 1. HumanEval
	Description: A set of 164 Python programming problems from OpenAI's HumanEval benchmark.
	Metrics: Pass@k (k=1, 10, 100)
	Expected Range: 70-80% pass@1 (based on Qwen2.5-Coder-32B baseline of ~76.8%)
	Status: Scheduled for post-training evaluation

	### 2. MBPP (Mostly Basic Python Programming)
	Description: 500 Python function synthesis problems from Google's MBPP dataset.
	Metrics: Pass@1, execution accuracy
	Expected Range: 80-85% pass@1 (based on Qwen2.5-Coder-32B baseline of ~82.3%)
	Status: Scheduled for post-training evaluation

	### 3. SWE-bench
	Description: Real-world GitHub issues requiring code modifications and debugging. This is the most challenging software engineering benchmark.
	Metrics: Resolution rate, edit similarity, test pass rate
	Expected Range: 15-25% resolution rate (based on similar 32B parameter models)
	Status: Planned for comprehensive testing post-training

	### 4. Tool Use Accuracy (Custom OpenClaw Suite)
	Description: 500 tasks covering OpenClaw-specific tool patterns: file operations, search, API calls, system commands, data processing, and multi-step workflows.
	Metrics: Task completion rate, tool call accuracy, parameter correctness, workflow success
	Expected Range: 85-92% overall task completion (conservative estimate based on fine-tuning for tool patterns)
	Status: Evaluation framework in development

	## Additional Evaluations

	### Context Understanding
	- Long-context benchmark: Testing 128K token window utilization
	- Multi-file reasoning: Cross-file code comprehension and modification

	### Specialized Domains
	- Voice Integration: Voice command processing and response generation
	- Documentation Generation: Quality assessment of auto-generated API docs
	- Code Review: Bug detection and suggestion quality

	## Results Template

	Once evaluations are complete, results will be published in the following format:

	\| Benchmark \| Pass@1 / Score \| Sample Size \| Evaluation Date \| Notes \|
	\|-----------\|----------------\|-------------\|-----------------\|-------\|
	\| HumanEval \| TBD \| 164 problems \| TBD \| Standard Python coding \|
	\| MBPP \| TBD \| 500 problems \| TBD \| Basic Python synthesis \|
	\| SWE-bench \| TBD \| Varies \| TBD \| Real-world GitHub issues \|
	\| Tool Use \| TBD \| 500 tasks \| TBD \| OpenClaw tool patterns \|
	\| GSM8K \| TBD \| 1319 problems \| TBD \| Math reasoning (optional) \|

	## Benchmark Methodology

	### Testing Conditions
	- Temperature: 0.2 (for code generation tasks)
	- Top_p: 0.95
	- Batch size: 1 (unless otherwise noted)
	- Hardware: NVIDIA A100 80GB (or equivalent)
	- Quantization: AWQ 4-bit where applicable
	- Inference engine: vLLM or similar for throughput testing

	### Evaluation Process
	1. Preprocessing: Standardized test set preparation with sanitization
	2. Inference: Automated generation of responses for each test case
	3. Verification: Automated test execution for coding problems
	4. Analysis: Statistical aggregation and result compilation
	5. Documentation: Detailed methodology and raw results publication

	## Timeline

	- Training Completion: [Date to be announced]
	- Benchmark Execution: 1-2 weeks post-training
	- Results Analysis: 1 week
	- Public Release: 1 week after analysis completion

	## Publication

	Results will be published in multiple formats:

	1. This document (BENCHMARKS.md) - Summary tables and key findings
	2. Detailed report ( BENCHMARKS_DETAILED.md) - In-depth methodology and raw scores
	3. GitHub Release - Official results with reproducible evaluation scripts
	4. OpenRouter listing - Performance metrics for model comparison

	---

	Stack 2.9 Benchmark Status: In Progress \| Results Coming Soon

	# Benchmark Results - Stack 2.9

	> Note: These benchmarks are currently in progress. Results will be published after training is complete.

	## Benchmark Overview

	Stack 2.9 will be evaluated on a comprehensive suite of benchmarks to measure coding capabilities, tool use proficiency, and overall model performance. The evaluation framework includes both standard coding benchmarks and custom tool-use scenarios.

	## Planned Benchmarks

	### 1. HumanEval
	Description: A set of 164 Python programming problems from OpenAI's HumanEval benchmark.
	Metrics: Pass@k (k=1, 10, 100)
	Expected Range: 70-80% pass@1 (based on Qwen2.5-Coder-32B baseline of ~76.8%)
	Status: Scheduled for post-training evaluation

	### 2. MBPP (Mostly Basic Python Programming)
	Description: 500 Python function synthesis problems from Google's MBPP dataset.
	Metrics: Pass@1, execution accuracy
	Expected Range: 80-85% pass@1 (based on Qwen2.5-Coder-32B baseline of ~82.3%)
	Status: Scheduled for post-training evaluation

	### 3. SWE-bench
	Description: Real-world GitHub issues requiring code modifications and debugging. This is the most challenging software engineering benchmark.
	Metrics: Resolution rate, edit similarity, test pass rate
	Expected Range: 15-25% resolution rate (based on similar 32B parameter models)
	Status: Planned for comprehensive testing post-training

	### 4. Tool Use Accuracy (Custom OpenClaw Suite)
	Description: 500 tasks covering OpenClaw-specific tool patterns: file operations, search, API calls, system commands, data processing, and multi-step workflows.
	Metrics: Task completion rate, tool call accuracy, parameter correctness, workflow success
	Expected Range: 85-92% overall task completion (conservative estimate based on fine-tuning for tool patterns)
	Status: Evaluation framework in development

	## Additional Evaluations

	### Context Understanding
	- Long-context benchmark: Testing 128K token window utilization
	- Multi-file reasoning: Cross-file code comprehension and modification

	### Specialized Domains
	- Voice Integration: Voice command processing and response generation
	- Documentation Generation: Quality assessment of auto-generated API docs
	- Code Review: Bug detection and suggestion quality

	## Results Template

	Once evaluations are complete, results will be published in the following format:

	\| Benchmark \| Pass@1 / Score \| Sample Size \| Evaluation Date \| Notes \|
	\|-----------\|----------------\|-------------\|-----------------\|-------\|
	\| HumanEval \| TBD \| 164 problems \| TBD \| Standard Python coding \|
	\| MBPP \| TBD \| 500 problems \| TBD \| Basic Python synthesis \|
	\| SWE-bench \| TBD \| Varies \| TBD \| Real-world GitHub issues \|
	\| Tool Use \| TBD \| 500 tasks \| TBD \| OpenClaw tool patterns \|
	\| GSM8K \| TBD \| 1319 problems \| TBD \| Math reasoning (optional) \|

	## Benchmark Methodology

	### Testing Conditions
	- Temperature: 0.2 (for code generation tasks)
	- Top_p: 0.95
	- Batch size: 1 (unless otherwise noted)
	- Hardware: NVIDIA A100 80GB (or equivalent)
	- Quantization: AWQ 4-bit where applicable
	- Inference engine: vLLM or similar for throughput testing

	### Evaluation Process
	1. Preprocessing: Standardized test set preparation with sanitization
	2. Inference: Automated generation of responses for each test case
	3. Verification: Automated test execution for coding problems
	4. Analysis: Statistical aggregation and result compilation
	5. Documentation: Detailed methodology and raw results publication

	## Timeline

	- Training Completion: [Date to be announced]
	- Benchmark Execution: 1-2 weeks post-training
	- Results Analysis: 1 week
	- Public Release: 1 week after analysis completion

	## Publication

	Results will be published in multiple formats:

	1. This document (BENCHMARKS.md) - Summary tables and key findings
	2. Detailed report ( BENCHMARKS_DETAILED.md) - In-depth methodology and raw scores
	3. GitHub Release - Official results with reproducible evaluation scripts
	4. OpenRouter listing - Performance metrics for model comparison

	---

	Stack 2.9 Benchmark Status: In Progress \| Results Coming Soon