datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
perfectblend-qwen3-4b-regen
perfectblend-qwen3-4b-regen
Single-turn SFT-style corpus for training speculative-decoding draft models (DFlash/MTP-style) against Qwen/Qwen3-4B as the target. Prompts come from an open-perfectblend-derived blend; every assistant response was regenerated by Qwen3-4B itself, so the token distribution matches the target model exactly. The sampled output_token_ids are included, letting trainers supervise on the target's own decode without re-tokenization drift.… See the full description on the dataset page: https://huggingface.co/datasets/susun-123/perfectblend-qwen3-4b-regen.Code-170k-susu
Dataset Description
Code-170k-susu is a groundbreaking dataset containing 176,999 programming conversations, originally sourced from glaiveai/glaive-code-assistant-v2 and translated into Susu, making coding education accessible to Susu speakers.
🌟 Key Features
176,999 high-quality conversations about programming and coding
Pure Susu language - democratizing coding education
Multi-turn dialogues covering various programming concepts
Diverse topics: algorithms, data… See the full description on the dataset page: https://huggingface.co/datasets/michsethowusu/Code-170k-susu.v7_socratic_data
v7 Socratic Grade 10 science chats (socratic_v7_final_v3)
JSONL of messages (system / user / assistant) for QLoRA SFT of a Socratic Grade 10 science tutor.
File: socratic_v7_final_v3.jsonl (~1429 chats).
Train 4B adapters to Susu11/v7_4b_qwen (do not overwrite Susu11/v9socratic4b).
Train 7B adapters to Susu11/v7_qwen7b (do not overwrite Susu11/qwen2.5-7b-socratic-tutor).
Code: Sushey01/Socratic-Model-Fine-Tune. Upload: python upload_v7_socratic_data.py.