Instructions to use SEE7/Astheral with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use SEE7/Astheral with Transformers:

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

pipe = pipeline("text-generation", model="SEE7/Astheral")
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("SEE7/Astheral")
model = AutoModelForCausalLM.from_pretrained("SEE7/Astheral", 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 SEE7/Astheral with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "SEE7/Astheral"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SEE7/Astheral",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/SEE7/Astheral

SGLang

How to use SEE7/Astheral 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 "SEE7/Astheral" \
    --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": "SEE7/Astheral",
		"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 "SEE7/Astheral" \
        --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": "SEE7/Astheral",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use SEE7/Astheral with Docker Model Runner:
```
docker model run hf.co/SEE7/Astheral
```

SEE7 commited on 2 days ago

Commit

0ab960c

verified ·

1 Parent(s): a5093c6

Document fixed-budget evaluation limits and regression; verify generated coding sample

Browse files

Files changed (3) hide show

README.md +6 -0
artifact_manifest.json +4 -0
coding_smoke_test.json +7 -0

README.md CHANGED Viewed

@@ -70,3 +70,9 @@ These narrow checks do not establish general superiority. Full prompts and outpu
 Use the --system option to replace the default assistant instructions. Example: python inference.py "Explain a Python generator." --mode instruct --system "Answer concisely and use one practical example."
 Thinking mode appends the upstream mode instruction to your system instructions. Custom system instructions influence generation; they do not erase behaviors learned into the weights or change the upstream license.

 Use the --system option to replace the default assistant instructions. Example: python inference.py "Explain a Python generator." --mode instruct --system "Answer concisely and use one practical example."
 Thinking mode appends the upstream mode instruction to your system instructions. Custom system instructions influence generation; they do not erase behaviors learned into the weights or change the upstream license.
+## Evaluation limits and regression
+Generation comparisons used identical per-mode caps: 256 new tokens for instruct and 512 for thinking. Two base-model thinking answers hit that budget before finishing, so 2/8 versus 6/8 measures the specific checks within that budget, not a general reasoning gain. Astheral regressed on the grouped-counting problem: it returned 9 instead of 18. It also failed the two-line formatting case. These failures remain in the published outputs.
+The generated median-function sample passed six remote functional checks (odd/even length, negative values, singleton, duplicates, and empty-input error). This does not establish general coding superiority.

artifact_manifest.json CHANGED Viewed

@@ -142,5 +142,9 @@
   "validation.jsonl": {
     "bytes": 21845,
     "sha256": "ad6b95669ef5cd1572feee70021e70d54a112a52180205aadc7cde93c2ebe70d"
   }
 }

   "validation.jsonl": {
     "bytes": 21845,
     "sha256": "ad6b95669ef5cd1572feee70021e70d54a112a52180205aadc7cde93c2ebe70d"
+  },
+  "coding_smoke_test.json": {
+    "bytes": 243,
+    "sha256": "f875753bee6c765a98355d1ecb715e7c608e6c48960c26e8041cef6d79b4e273"
   }
 }

coding_smoke_test.json ADDED Viewed

	@@ -0,0 +1,7 @@

+{
+  "sample": "coding median function in astheral_outputs.json",
+  "checks_passed": 6,
+  "checks_total": 6,
+  "scope": "Only this generated sample was executed; not a general coding benchmark.",
+  "execution_location": "Hugging Face CPU Job"
+}