---

license: apache-2.0

language:

• en

pipeline_tag: text-generation

tags:

• gguf
• llama.cpp
• causal-lm
• gpt2
• educational
• from-scratch
• cascamini

base_model: beardymcgee/cascamini-gpt2-210m-edu-5b

---

CascaMini-GPT2-210M-Edu-5B GGUF

This repository contains GGUF quantised files for CascaMini-GPT2-210M-Edu-5B.

The original Transformers model is available here:

beardymcgee/cascamini-gpt2-210m-edu-5b

Model details

• Model family: CascaMini
• Architecture: GPT-2-style decoder-only causal language model
• Parameters: approximately 209.5M
• Context length: 1024 tokens
• Tokenizer: GPT-2 BPE
• Training tokens: approximately 5.24B
• Best validation loss: 3.2245991230010986
• Model type: base text-completion model
• Instruction tuned: no
• Chat model: no

Available files

| File | Type |

|---|---|

| cascamini-gpt2-210m-edu-5b-q8_0.gguf | Q8_0 quantised GGUF |

llama.cpp usage

Use llama-completion, not llama-cli, because this is a base completion model and not a chat model.

llama-completion \
  -m cascamini-gpt2-210m-edu-5b-q8_0.gguf \
  -p "The main purpose of an operating system is" \
  -n 160 \
  --temp 0.55 \
  --top-k 50 \
  --top-p 0.9 \
  --repeat-penalty 1.1

Known limitations

This model is a small educational base model trained from scratch as a learning project.

It can produce:

• factual errors
• circular explanations
• repetition
• invented details
• weak reasoning
• poor instruction-following behaviour

It should not be used as a factual tutor or assistant without additional verification.