Downloads · 30 days
41
73% of all-time downloads
minimind-fr-creative
yassinsiouda/minimind-fr-creative
minimind-fr-creative is a machine learning model from yassinsiouda. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for pytorch. The card lists the license as apache-2.0.
French creative-writing specialist. ~64 M params (hiddensize=768, numhiddenlayers=8, dense), MiniMind architecture. SFT from minimind-fr-creative's base (pretrain-enfr - base SFT - agentic SFT).
All-time downloads
56
Public
Repo size
266 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pth138 MB · 52%
At a glance
- Library
- pytorch
- License
- apache-2.0
- Access
- Public
- Created
- Sep 3, 2026
- Updated
- Sep 3, 2026
- SHA
- 8f84fc10
Base models
Model card
From the Hugging Face model README
minimind-fr-creative
French creative-writing specialist. ~64 M params (hidden_size=768, num_hidden_layers=8, dense),
MiniMind architecture. SFT from minimind-fr-creative's base (pretrain-enfr -> base SFT -> agentic SFT).
Round-1 prototype — fluent but small; treat outputs accordingly. Training framework: https://github.com/jingyaogong/minimind.
Files
| file | format |
|---|---|
spec-creative_768.pth | raw PyTorch state_dict (fp16) — load with MiniMindForCausalLM(strict=False) |
spec-creative-f16.gguf | GGUF F16 — exported via Qwen3ForCausalLM, runs in llama.cpp / Ollama / LM Studio |
tokenizer.json, tokenizer_config.json | byte-level BPE, vocab_size=6400, EN/FR |
Config: num_attention_heads 8, num_key_value_heads 4, vocab_size 6400,
max_position_embeddings 32768, rope_theta 1e6, tied embeddings, no MoE.
Training data
| dataset | contribution |
|---|---|
yassinsiouda/minimind-fr-creative-data | packaged training file for this model |
angeluriot/french_instruct | creative-instruction rows carved by keyword regex (54,644) |
jpacifico/French-Alpaca-dataset-Instruct-110K | creative rows (22,826) |
PleIAs/French-PD-Books | Continue ce texte continuation pairs (8,000) |
allenai/tulu-3-sft-mixture | base-SFT replay (anti-forgetting) |
NousResearch/hermes-function-calling-v1 | agentic replay |
nvidia/Nemotron-SFT-SWE-v3.5 | terminal-loop replay |
135,078 rows. French-native, no translation. ~37.5% replay of the base SFT mix (60% general / 40% agentic) limits catastrophic forgetting; ~15% of rows carry a short planning <think>. Recipe: convert_spec_creative.py.
Run (GGUF)
llama-cli -m spec-creative-f16.gguf -p "..." -ngl 99
Limitations
64 M parameters; English domain data for the specialists (French comes from the
bilingual base, so domain idiom is anglicised); agentic capability is SFT-only
(short 2-4 step tool loops); safety limited to the router's thin unsafe bucket.
License
Apache-2.0 (weights). Upstream dataset licenses govern downstream use — see the dataset repo.
- Library
- pytorch
- License
- apache-2.0
- Languages
- fr, en
- Created
- Sep 3, 2026
- Updated
- Sep 3, 2026