taiyi-lab

toolathlon-qwen36-27b-redteam5k-reasoning-sft-20260628-epoch3

Deploy Dedicated

README

License: other

Identity

HF repo: taiyi-lab/toolathlon-qwen36-27b-redteam5k-reasoning-sft-20260628-epoch3
Epoch: 3/3
Checkpoint step: 3750
Local checkpoint path at upload time: /data/train/toolathlon_qwen36_27b/output/redteam5k_reasoning_3epoch_container_20260628_155003/redteam5k_reasoning_full_3epoch/v0-20260628-155034/checkpoint-3750
Local run dir: /data/train/toolathlon_qwen36_27b/output/redteam5k_reasoning_3epoch_container_20260628_155003/redteam5k_reasoning_full_3epoch/v0-20260628-155034
Uploaded at: 2026-06-29 09:53:00 UTC
Visibility at upload: public

Base Model

Local base model: /data/models/qwen3.6-27B
Local Megatron/MCore model: /data/models/qwen3.6-27B-mcore

Training Data

Source dataset: taiyi-lab/toolathlon-gym-redteam-5k-sft-20260627
Split used: reasoning_sft
Converted local training file: /data/train/toolathlon_qwen36_27b/data/redteam5k_reasoning_train_192k.jsonl
Manifest: /data/train/toolathlon_qwen36_27b/data/manifest_redteam5k_reasoning_192k.csv
Records kept/excluded: 5000 / 0
Max length: 196608
Token length p50/p90/p99/max: 19221.0 / 49172.0 / 65874.0 /

Training Configuration

Framework: MS-Swift / Megatron, inside Docker container swift
Tuning mode: full fine-tune
Precision: bf16
Tensor parallel size: 8
Pipeline parallel size: 1
Context parallel size: 1
Max length: 196,608
Save strategy: epoch
Optimizer/RNG states: not saved
Launch log: /data/train/toolathlon_qwen36_27b/logs/redteam5k_reasoning_3epoch_container_20260628_155003.log

Selected args:

json
{
  "model": "/data/models/qwen3.6-27B",
  "mcore_model": "/data/models/qwen3.6-27B-mcore",
  "dataset": [
    "/data/train/toolathlon_qwen36_27b/data/redteam5k_reasoning_train_192k.jsonl"
  ],
  "output_dir": "/data/train/toolathlon_qwen36_27b/output/redteam5k_reasoning_3epoch_container_20260628_155003/redteam5k_reasoning_full_3epoch/v0-20260628-155034",
  "num_train_epochs": 3,
  "max_length": 196608,
  "truncation_strategy": "delete",
  "agent_template": "qwen3_5",
  "loss_scale": "default",
  "tuner_type": "full",
  "lr": 3e-06,
  "min_lr": 3e-07,
  "lr_warmup_fraction": 0.05,
  "weight_decay": 0.1,
  "micro_batch_size": 2,
  "global_batch_size": 4,
  "tensor_model_parallel_size": 8,
  "pipeline_model_parallel_size": 1,
  "sequence_parallel": true,
  "bf16": true,
  "save_strategy": "epoch",
  "no_save_optim": true,
  "no_save_rng": true
}

Checkpoint Contents

Total size: 50.97 GiB
Safetensor shards: 12
Shard names: model-00001-of-00012.safetensors, model-00002-of-00012.safetensors, model-00003-of-00012.safetensors, model-00004-of-00012.safetensors, model-00005-of-00012.safetensors, model-00006-of-00012.safetensors, model-00007-of-00012.safetensors, model-00008-of-00012.safetensors, model-00009-of-00012.safetensors, model-00010-of-00012.safetensors, model-00011-of-00012.safetensors, model-00012-of-00012.safetensors

Intended Use

This checkpoint is intended for Toolathlon-style self-hosted tool-use evaluation and research comparisons against the base Qwen3.6 27B model and adjacent epochs.

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Model Details

Model Provider

taiyi-lab

Model Tree

Base

this model

Input Modalities

TextImageVideo

Output Modalities

Text

Supported Functionality