Instructions to use beardymcgee/CascaMini-350M-C24-CoreMix-v1-15B-VO-Muon-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use beardymcgee/CascaMini-350M-C24-CoreMix-v1-15B-VO-Muon-v1 with Transformers:

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

pipe = pipeline("text-generation", model="beardymcgee/CascaMini-350M-C24-CoreMix-v1-15B-VO-Muon-v1")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("beardymcgee/CascaMini-350M-C24-CoreMix-v1-15B-VO-Muon-v1")
model = AutoModelForCausalLM.from_pretrained("beardymcgee/CascaMini-350M-C24-CoreMix-v1-15B-VO-Muon-v1", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use beardymcgee/CascaMini-350M-C24-CoreMix-v1-15B-VO-Muon-v1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "beardymcgee/CascaMini-350M-C24-CoreMix-v1-15B-VO-Muon-v1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "beardymcgee/CascaMini-350M-C24-CoreMix-v1-15B-VO-Muon-v1",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/beardymcgee/CascaMini-350M-C24-CoreMix-v1-15B-VO-Muon-v1

SGLang

How to use beardymcgee/CascaMini-350M-C24-CoreMix-v1-15B-VO-Muon-v1 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 "beardymcgee/CascaMini-350M-C24-CoreMix-v1-15B-VO-Muon-v1" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "beardymcgee/CascaMini-350M-C24-CoreMix-v1-15B-VO-Muon-v1",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "beardymcgee/CascaMini-350M-C24-CoreMix-v1-15B-VO-Muon-v1" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "beardymcgee/CascaMini-350M-C24-CoreMix-v1-15B-VO-Muon-v1",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use beardymcgee/CascaMini-350M-C24-CoreMix-v1-15B-VO-Muon-v1 with Docker Model Runner:
```
docker model run hf.co/beardymcgee/CascaMini-350M-C24-CoreMix-v1-15B-VO-Muon-v1
```

CascaMini-350M-C24-CoreMix-v1-15B-VO-Muon-v1 / TRAINING_DATA_AND_ATTRIBUTION.md

beardymcgee

Finalize public release documentation: TRAINING_DATA_AND_ATTRIBUTION.md

8e32224 verified 21 days ago

preview code

Raw

History Blame Contribute Delete

5.43 kB

Training Data and Attribution

This file documents the training-data provenance retained for CascaMini-350M-C24-CoreMix-v1-15B-VO-Muon-v1.

It is a transparency record, not legal advice and not a replacement for the terms of the original datasets or source works.

Mixture

Source family	M3 weight	Materialised source
FineWeb-Edu	55.00%	HuggingFaceFW/fineweb-edu
Wikimedia	10.00%	common-pile/wikimedia_filtered
peS2O	8.00%	common-pile/peS2o_filtered
arXiv	5.00%	common-pile/arxiv_papers_filtered
Stack v2 code	12.00%	common-pile/stackv2_edu_filtered
Math StackExchange	5.00%	common-pile/stackexchange_filtered
LibreTexts	0.75%	common-pile/libretexts_filtered
Pressbooks	2.00%	common-pile/pressbooks_filtered
OER Commons	0.75%	common-pile/oercommons_filtered
DOAB	0.50%	common-pile/doab_filtered
Structured Stack v2	1.00%	common-pile/stackv2_edu_filtered

Frozen source revisions retained in the canonical corpus

Source	Repository	Revision
Wikimedia	common-pile/wikimedia_filtered	`0641bb84bd9b7162bcddf8be7836822161a9a342`
peS2O	common-pile/peS2o_filtered	`297747513bfb0ff1fbf61ddad3b03319d0f04597`
arXiv	common-pile/arxiv_papers_filtered	`033cf7f53f9b348deec868c1a5a48484f3ee9e52`
Stack v2	common-pile/stackv2_edu_filtered	`c354dbe88469a1153e97c6a63ac50591849654de`
StackExchange	common-pile/stackexchange_filtered	`c0ac7373830c688a43fc12d1988c4b19ccd884ab`
LibreTexts	common-pile/libretexts_filtered	`70388bca52b4a93515e14b1d56618fd7944988fd`
Pressbooks	common-pile/pressbooks_filtered	`1a1d3b50d77f834370f8eb4c0d174668dd1676bb`
OER Commons	common-pile/oercommons_filtered	`506b6159dadcbc0dc67611cea024eedb04232fb2`
DOAB	common-pile/doab_filtered	`defb24ca72ef6aba6ce0228b669eec06dcfbffbc`

FineWeb-Edu was trained from the project's frozen local stream. The upstream dataset family is HuggingFaceFW/fineweb-edu; the project's immutable source stream identity is retained separately in the CascaMini release documentation.

Retained licence-evidence audit

A corrected publication audit scanned all canonical CoreMix training JSONLs.

Every one of the 6,699,937 CoreMix training records had:

licence_evidence_mode;
licence_fields;
non-empty licence_fields.

The retained evidence modes were record_metadata.

Wikimedia

1,689,491 records.

Retained metadata: 100% CC BY-SA 4.0.

peS2O

200,894 records.

Retained metadata distribution:

97.255% CCBY;
2.011% CCBYSA;
0.559% CC0;
0.168% public-domain marker pd;
0.007% public-domain.

arXiv papers

61,817 records.

Retained metadata distribution:

81.478% CC BY 4.0;
8.326% CC BY 3.0;
4.178% CC BY-SA 4.0;
3.255% CC0;
2.763% public domain.

Math StackExchange

1,408,077 records.

Retained metadata distribution:

51.613% CC BY-SA 4.0;
47.367% CC BY-SA 3.0;
1.021% CC BY-SA 2.5.

Source-author metadata was non-empty for approximately 85.2% of these records; source URLs/provenance were retained separately by the materialisation pipeline.

LibreTexts

2,969 records.

Retained metadata included:

53.520% CC BY 4.0;
24.251% CC BY-SA 4.0;
13.405% GNU Free Documentation License;
6.096% CC BY 3.0;
2.593% CC BY-SA 3.0;
0.135% public domain.

Pressbooks

49,327 records.

Retained metadata:

77.244% CC BY 4.0;
14.422% CC BY-SA 4.0;
5.291% public domain;
3.043% CC0.

OER Commons

3,565 records.

Retained metadata:

88.219% CC BY 4.0;
7.686% CC BY-SA 4.0;
4.095% public domain.

DOAB

203,580 records.

Retained metadata:

59.698% CC BY 3.0;
40.283% CC BY 4.0;
0.018% CC BY-SA 4.0.

Stack v2 code and structured code/text

The canonical filtered Stack v2 records had 100% non-empty per-record licence_fields.

The largest detected licence families were MIT, Apache-2.0, BSD-3-Clause, BSD-2-Clause, Unlicense, BSL-1.0, CC0-1.0, Zlib, ISC, and other permissive SPDX licences, including multi-licence records.

The Stack v2 upstream terms state that source repositories carry varying licences and that original licence/attribution obligations remain relevant.

FineWeb-Edu

FineWeb-Edu is published on the Hugging Face Hub under ODC-BY metadata. The project's frozen pretokenised training stream does not itself retain per-document licence fields, so upstream FineWeb-Edu/Common Crawl provenance should be consulted for source-level terms.

Important downstream note

This repository distributes model artifacts, not copies of the training datasets.

The relationship between third-party training-data licences and model weights or generated outputs can depend on jurisdiction and facts. This release does not claim that all source licences collapse into one licence for the model.

If a generated output reproduces protected source material, users should consider the rights and obligations applicable to that material.

Upstream references

FineWeb-Edu: https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu
Common Pile filtered data: https://huggingface.co/collections/common-pile/common-pile-v01-filtered-data
Stack v2: https://huggingface.co/datasets/bigcode/the-stack-v2
StackExchange licensing: https://stackoverflow.com/help/licensing
Wikimedia terms: https://foundation.wikimedia.org/wiki/Policy:Terms_of_Use