datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
gemma-3-taide-12b-chat-eval-logs-and-scoresgemma-3-4b-it-eval-logs-and-scoresgemma-3-4B-T1-it-eval-logs-and-scoresGemma-3-12b-it-eval-logs-and-scoresgemma-3-27b-it-eval-logs-and-scoresgemma3n-conversational-reasoning
Gemma3N Conversational Reasoning
This dataset is prepared for Unsloth Gemma3/Gemma3N conversational notebooks that use:
from datasets import load_dataset
from unsloth.chat_templates import standardize_data_formats
dataset = load_dataset("Cyleux/gemma3n-conversational-reasoning", split="train[:3000]")
dataset = standardize_data_formats(dataset)
Schema:
conversations: ShareGPT-style list of turns with from and value
metadata columns are included for analysis and filtering
Notes:… See the full description on the dataset page: https://huggingface.co/datasets/Cyleux/gemma3n-conversational-reasoning.gemma3-reasoning-dropin
Gemma3 Reasoning Drop-In Dataset
Drop-in replacement dataset for Gemma3-style notebooks expecting task, input, expected_output fields.
Usage
from datasets import load_dataset
dataset = load_dataset("Cyleux/gemma3-reasoning-dropin", split="train[:10000]")
Then your existing mapping code can stay the same:
task -> system
input -> user
expected_output -> assistant
Reasoning is preserved in expected_output using <think>...</think> blocks.
Stats
{
"input":… See the full description on the dataset page: https://huggingface.co/datasets/Cyleux/gemma3-reasoning-dropin.gemma3-reasoning-dropin-context
Gemma3 Reasoning Drop-in (Context Preserved)
Drop-in dataset with task/input/expected_output where input includes prior turns so follow-up replies remain coherent.
Use:
from datasets import load_dataset
dataset = load_dataset("Cyleux/gemma3-reasoning-dropin-context", split="train[:10000]")
Stats:
{
"input": "data/functiongemma_upload/train.jsonl",
"output": "data/gemma3_reasoning_dropin_context/train.jsonl",
"rows": 2225,
"samples_total": 1214,
"rows_written": 2225… See the full description on the dataset page: https://huggingface.co/datasets/Cyleux/gemma3-reasoning-dropin-context.gemma-3-270m-vismem-rag-opus-reasoning-1-vismem-kb-0515-1236
VisMem Knowledge Base (2326 entries, 1984x1984px)
Load with:
import requests
from vismem_core import VisMem
data = requests.get(
"https://huggingface.co/datasets/broadfield-dev/gemma-3-270m-vismem-rag-opus-reasoning-1-vismem-kb-0515-1236/resolve/main/vismem.png",
headers={"Authorization": "Bearer <TOKEN>"}).content
mem = VisMem.from_png_bytes(data)
results = mem.search(your_embedding, k=3)
gemma3n-conversational-reasoning-with-tools
Gemma3N Conversational Reasoning With Embedded Tool Traces
Prepared for Unsloth Gemma3/Gemma3N conversational notebooks that expect ShareGPT conversations.
Multi-turn conversations are preserved.
Reasoning blocks (<think>...</think>) are preserved.
Tool call traces are preserved by embedding them in assistant text as tags:
<tool_call ...>...</tool_call>
<tool_response ...>...</tool_response>
Use:
from datasets import load_dataset
from unsloth.chat_templates import… See the full description on the dataset page: https://huggingface.co/datasets/Cyleux/gemma3n-conversational-reasoning-with-tools.gemma-3-12b-it-lmsys-onpolicy-rollouts
On-policy chat rollouts: google/gemma-3-12b-it on LMSYS-Chat-1M prompts
Each row is a first-user-turn prompt sampled from
lmsys/lmsys-chat-1m and a
response generated on-policy by google/gemma-3-12b-it with vLLM (do_sample,
temperature 0.7, top_p 1.0, max_new_tokens 768, seed 42). 24,991 rows. Built to match
GemmaScope 2's instruction-tuned SAE training distribution (real model rollouts) for a
short KL+MSE ("end-to-end") finetune of the released GemmaScope-2 residual SAE.… See the full description on the dataset page: https://huggingface.co/datasets/iarcuschin/gemma-3-12b-it-lmsys-onpolicy-rollouts.gemma3n-conversational-reasoning-toolloop
Gemma3N Conversational Reasoning Tool-Loop
Gemma3N conversational dataset that preserves tool traces while avoiding training targets on tool responses.
Encoding:
Assistant emits tool calls: <tool_call ...>...</tool_call>
Tool outputs are user-side turns: <tool_response ...>...</tool_response>
This works with train_on_responses_only because user-side tool responses are masked from loss.
Use:
from datasets import load_dataset
from unsloth.chat_templates import… See the full description on the dataset page: https://huggingface.co/datasets/Cyleux/gemma3n-conversational-reasoning-toolloop.gemma-3-270m-opus-reasoning-1-vismem-kb-0516-0810
VisMem Knowledge Base (2326 entries, 1984x1984px)
Load with:
import requests
from vismem_core import VisMem
data = requests.get(
"https://huggingface.co/datasets/broadfield-dev/gemma-3-270m-opus-reasoning-1-vismem-kb-0516-0810/resolve/main/vismem.png",
headers={"Authorization": "Bearer <TOKEN>"}).content
mem = VisMem.from_png_bytes(data)
results = mem.search(your_embedding, k=3)