datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
to-tool-call-datasets-LFM2.5-pythonic
to-tool-call-datasets → LFM2.5 Pythonic tool-call format
A derivative of zhangdw/to-tool-call-datasets (apache-2.0)
normalized for fine-tuning Liquid AI LFM2 / LFM2.5 models, whose native tool-call format is Pythonic:
<|im_start|>assistant
<|tool_call_start|>[get_weather(location='Paris, France', unit='celsius')]<|tool_call_end|><|im_end|>
Nine public tool-calling corpora (APIGen-MT, ButtonInstruct, Glaive v2, GraphSyn, LoopTool, τ-bench train, ToolACE, When2Call, xLAM-60k)… See the full description on the dataset page: https://huggingface.co/datasets/hungbenjamin402/to-tool-call-datasets-LFM2.5-pythonic.scriber-lfm2.5-350m-polishing-de-training-v1
Scriber LFM2.5 German STT post-processing data
This repository contains the exact 2,000 German source/target pairs used to
train the final Scriber LFM2.5 350M local post-processing model.
The matching model is
Buttermilk03/scriber-lfm2.5-350m-polishing-de-qad-v1.
The complete production recipe and the lessons that determined it are in
TRAINING.md; machine-readable settings are in
training_recipe.json.
Data
Each JSONL row contains:
source: flat German… See the full description on the dataset page: https://huggingface.co/datasets/Buttermilk03/scriber-lfm2.5-350m-polishing-de-training-v1.Nemotron-Agentic-v1-LFM2.5-pythonic
Nemotron-Agentic-v1 → LFM2.5 Pythonic tool-call format
A derivative of nvidia/Nemotron-Agentic-v1 (cc-by-4.0)
normalized for fine-tuning Liquid AI LFM2 / LFM2.5 models, whose native tool-call format is Pythonic:
<|im_start|>assistant
<|tool_call_start|>[get_weather(location='Paris, France', unit='celsius')]<|tool_call_end|><|im_end|>
Multi-turn conversational tool-use trajectories (interactive_agent: goal decomposition with persona-seeded users; tool_calling: general function… See the full description on the dataset page: https://huggingface.co/datasets/hungbenjamin402/Nemotron-Agentic-v1-LFM2.5-pythonic.Nemotron-RL-IF-Gym-LFM2.5-prompts
Nemotron-RL instruction-following & abstention Gym tasks → LFM2.5 prompt format
Verifiers are NVIDIA NeMo-Gym rule checkers, not IFEval. Each row keeps the full verifier metadata in extra (verifier regex/string-match specs, schema_str/schema_type, exp_cal_state, answer) so the checks can be re-implemented or run through NeMo Gym's resources servers. qa_abstention rows carry a per-row license of CC BY-SA 4.0 from the source.
A derivative of [five nvidia/Nemotron-RL-* NeMo-Gym… See the full description on the dataset page: https://huggingface.co/datasets/hungbenjamin402/Nemotron-RL-IF-Gym-LFM2.5-prompts.IF-multi-constraints-upto5-LFM2.5-prompts
IF_multi_constraints_upto5 → LFM2.5 prompt format (for RLVR / rejection sampling / DPO)
A derivative of allenai/IF_multi_constraints_upto5 (odc-by)
normalized for fine-tuning Liquid AI LFM2 / LFM2.5 models, whose native tool-call format is Pythonic:
<|im_start|>assistant
<|tool_call_start|>[get_weather(location='Paris, France', unit='celsius')]<|tool_call_end|><|im_end|>
Prompt-only rows (prompt_only = true): Tulu-SFT instructions with up to 5 verifiable constraints from IFEval… See the full description on the dataset page: https://huggingface.co/datasets/hungbenjamin402/IF-multi-constraints-upto5-LFM2.5-prompts.Nemotron-SFT-Agentic-v2-LFM2.5-pythonic-dryrun
[DRY RUN — 2,000 rows/split] Nemotron-SFT-Agentic-v2 → LFM2.5 Pythonic tool-call format
A derivative of nvidia/Nemotron-SFT-Agentic-v2
(CC-BY-4.0) normalized for supervised fine-tuning of Liquid AI LFM2 / LFM2.5 models, whose
native tool-call format is Pythonic:
<|im_start|>assistant
<|tool_call_start|>[get_weather(location='Paris, France', unit='celsius')]<|tool_call_end|><|im_end|>
Every row was rendered through the official LiquidAI/LFM2.5-VL-3B chat template (identical
to… See the full description on the dataset page: https://huggingface.co/datasets/hungbenjamin402/Nemotron-SFT-Agentic-v2-LFM2.5-pythonic-dryrun.noval-corp-training-corpus-lfm2
Noval-Corp Training Corpus (LFM2.5 round)
What's inside — The fused CPT corpus of the LFM2.5-1.2B round: 12.17M tokens — 84.3% OMC documents, 8.7% curated telemetry prose, 7.0% general web replay.
Where it comes from — orbital-mining-corpus-seeded + mars-logs-curated + FineWeb-Edu (pinned revision) for the anti-forgetting replay (Ibrahim 2024).
How it was built / modified — Token-budgeted mixing, row-level shuffle, exact + MinHash/LSH near-dedup (build_corpus.py). Validation =… See the full description on the dataset page: https://huggingface.co/datasets/atenareply/noval-corp-training-corpus-lfm2.IF-RLVR-IFEval-prompts-LFM2.5
IF-RLVR prompts with IFEval verifier spec → LFM2.5 prompt format
A derivative of [nvidia/Nemotron-RL-instruction_following + allenai/RLVR-IFeval (via sungyub/if-verl-unified)](https://huggingface.co/datasets/nvidia/Nemotron-RL-instruction_following + allenai/RLVR-IFeval (via sungyub/if-verl-unified)) (odc-by)
normalized for fine-tuning Liquid AI LFM2 / LFM2.5 models, whose native tool-call format is Pythonic:
<|im_start|>assistant
<|tool_call_start|>[get_weather(location='Paris… See the full description on the dataset page: https://huggingface.co/datasets/hungbenjamin402/IF-RLVR-IFEval-prompts-LFM2.5.IF-multi-constraints-upto5-SFT-LFM2.5
IF_multi_constraints_upto5_SFT → LFM2.5 chat format
A derivative of UniLu/IF_multi_constraints_upto5_SFT (odc-by)
normalized for fine-tuning Liquid AI LFM2 / LFM2.5 models, whose native tool-call format is Pythonic:
<|im_start|>assistant
<|tool_call_start|>[get_weather(location='Paris, France', unit='celsius')]<|tool_call_end|><|im_end|>
SFT-ready precise-instruction-following pairs: the allenai IF-RLVR prompts answered by Gemma-4-31B-it and filtered with the official IFBench… See the full description on the dataset page: https://huggingface.co/datasets/hungbenjamin402/IF-multi-constraints-upto5-SFT-LFM2.5.Nemotron-SFT-Agentic-v2-LFM2.5-pythonic
Nemotron-SFT-Agentic-v2 → LFM2.5 Pythonic tool-call format
A derivative of nvidia/Nemotron-SFT-Agentic-v2
(CC-BY-4.0) normalized for supervised fine-tuning of Liquid AI LFM2 / LFM2.5 models, whose
native tool-call format is Pythonic:
<|im_start|>assistant
<|tool_call_start|>[get_weather(location='Paris, France', unit='celsius')]<|tool_call_end|><|im_end|>
Every row was rendered through the official LiquidAI/LFM2.5-VL-3B chat template (identical
to the LFM2.5 text models'… See the full description on the dataset page: https://huggingface.co/datasets/hungbenjamin402/Nemotron-SFT-Agentic-v2-LFM2.5-pythonic.lfm2.5-1.2b-blindspots
LFM2.5-1.2B-Base Blind Spots Dataset
Model Link: https://huggingface.co/LiquidAI/LFM2.5-1.2B-BaseNotebook Link: https://colab.research.google.com/drive/13sDzKbvsaYxpKwBD-ePktBrDFX2qoPiI?usp=sharing
Model Loading and Inference
I loaded the model in a high-level manner using the pipeline function from the transformers library:
# Loading the model using the pipeline function from the transformers library
from transformers import pipeline
pipe = pipeline("text-generation"… See the full description on the dataset page: https://huggingface.co/datasets/Almaan/lfm2.5-1.2b-blindspots.HardGen-LFM2.5-pythonic
HardGen (FunReason-MT) → LFM2.5 Pythonic tool-call format
⚠️ Evaluation contamination notice. The source was generated by sampling in the Berkeley Function-Calling Leaderboard (BFCL) multi-turn environment (Gorilla file system, trading bot, etc.). Do not train on it if you report BFCL numbers; use it for analysis, as a hard held-out set, or with full awareness of the overlap.
A derivative of Bingguang/HardGen (apache-2.0)
normalized for fine-tuning Liquid AI LFM2 / LFM2.5… See the full description on the dataset page: https://huggingface.co/datasets/hungbenjamin402/HardGen-LFM2.5-pythonic.lfm25-blind-spots
🔍 Blind Spots — LiquidAI/LFM2.5-1.2B-Thinking
This dataset documents failure cases (blind spots) discovered while probing
LiquidAI/LFM2.5-1.2B-Thinking — a base pretrained reasoning model released
in early 2026, with ~1.17B parameters, part of LiquidAI's LFM2.5 family.
The model was selected because it satisfies all criteria:
✅ Released within the last 6 months (January–February 2026)
✅ Between 0.6B and 6B parameters (~1.17B)
✅ Base model — not fine-tuned for a specific… See the full description on the dataset page: https://huggingface.co/datasets/omarSaqr/lfm25-blind-spots.lfm2-24b-a2b-427xTrace of LFM2-24B-A2B LLM by LiquidAI.
Data count (Total: 427):
English - 211
Russian - 216
Data is presented in ShareGPT format and each conversation split by newline.
Brought to you by sapbot from Romarchive
lfm-2.5-1.2b-instruct-308xTrace of LFM2.5 1.2B Instruct LLM.
Data count (Total: 308):
English - 198
Russian - 110
Data is presented in {"messages":[{"role":"user", "content":"Prompt"}, {"role":"assistant", "content": "Response"}]} format and each conversation split by newline.
lfm25-base-failure-cases
Base Model Failure Cases Dataset
This dataset contains diverse failure cases from a base language model: inputs where the model’s output was incorrect or undesirable, along with the expected (correct or preferred) output. It is intended for analyzing blind spots and for fine-tuning or evaluation.
Model tested
LiquidAI/LFM2.5-1.2B-Base
Type: Base (pre-trained only) causal language model; no instruction tuning.
Parameters: 1.2B.
Released: January 2026 on Hugging Face… See the full description on the dataset page: https://huggingface.co/datasets/mihretgold/lfm25-base-failure-cases.lfm2.5-8b-a1b-506xTrace of LFM2.5 8B A1B LLM made by LiquidAI.
Data is presented in ChatML format and each conversation split by newline. Ready to be used for fine-tuning.
Example:
{"messages":[{"role":"user", "content":"Hello!"}, {"role":"assistant", "content":"Hello!"}]}
Brought to you by sapbot from Romarchive
lfm25-1.2b-base-blindspots
LFM2.5-1.2B-Base Blind Spots
This dataset documents 10 failure cases observed while testing the base model LiquidAI/LFM2.5-1.2B-Base.
Model Tested
Model: LiquidAI/LFM2.5-1.2B-Base
Link: https://huggingface.co/LiquidAI/LFM2.5-1.2B-Base
I selected this model because it is a base pretrained model rather than a checkpoint finetuned for a specific downstream application.
How I Loaded the Model
import torch
from transformers import AutoModelForCausalLM… See the full description on the dataset page: https://huggingface.co/datasets/progga002/lfm25-1.2b-base-blindspots.