CHUD-Qwen3.6-27B

CHUD-Qwen3.6-27B

ORPO run with synthetic data from Grok 4.5.

Like most of my models - untested

Training Configuration

Datasets

Trained on 3 concatenated datasets:

1. nbeerbower/weasel-dpo (split: train)
2. nbeerbower/seX-ai-dpo (split: train)
3. nbeerbower/grok-politically-incorrect-dpo (split: train)

Reproduce this training run

This model was trained with Merlina. Credentials are not included — Merlina will use your own HF_TOKEN and WANDB_API_KEY from .env or the form.

Paste this code into Merlina's Load Configuration → From Code to rebuild the exact training setup:

merlina-config-v1:H4sIAOf1bGoC_81WS4_bRgz-K4s5247t3Ww3e8u2hwLNHoI2QIGiGFASJU08r8zDjrvY_15ypLGdNEGAIIfqRHEoDh8fP-pJSIMJOkgg7p-EBYPiXvz867tflm8PaK9Xt8vtTw9iITqMbVA-KWfJ4I8Ayio7XLXO9mrIAfjgKo4QsLvqgzNXcPWIQSsLy8TGpDauQ70iXwmGKO7_-nshYjuioZuFmWxfpNnxcnIsqoncY4jl7s2iGp91YrvarDbieSEaiCjLRaS1DWJo3AHDiwc1POSmgc-ycjn5nOTX087kTrtAIaaQcSFYlkHcX29nGbQf6fT2Zn7vgvPkVdyvV2vONAyYOKCskXMW2Usf3HuuqDvYKg-QsMq7Knyowr4KbhKocLNHmZyMsMepmiWABHEn09GXhF6_-_31G_nmkb7VCIErK6lXdHaHy_XtQthsJHrXjhTclsuX2lFG9Q-W1yFAp9AmCW2bTdalyzIm9GROGRv4KDXaIY1kvr65mzQUo_HpdLB5eX1bAkZd44KcHONgbnZpGKld8KxukMG4XlGnNTT0VTTOpZEM56oOYEyxeLkQysCAMmB0Ok_g3Ky3tRkB7O6iV9zMrjSzBx2pm0Z18hREBONLk6ZO80hETDwV0eXQ4lmqaYx5GOjLHuhwIQLVUaruU9wdkJzoZVcSi14rcjjlze8pIJiS1xzQSSMvO7FZ00Pghq5TnCNQSUogjKjvCCrin0tQPyKohWip7sZKA96Xb57E1H7yOAtkM7qIPKazwGG9xzYhx3USn58X35HLENxu6R3loFrQ-rhUtnUhkMv_X340or0LBgqoJqlmmtwOLd0b6GO00GiqASF-V-6cAMlDIztM5O2zs-c6XjOPfZv3fmBW7MwSExPLYYw0jFHWJOdBwi5TD9pCO3PJiw6JSZiMhuPpWnm6K2FMcydo5okK4pF4x1RymRmjV1pLJMXxa3RC8UXkkG-2Ezsxp0lLEVKLmUwOEAyRctlgM-vgnidsYrn1ajttgZuGYTTHz4oD2K6pSfocRybjMZ9UtKWGaXnIBqkkKKkMDrp6PvZsTQmpfSnNpMWP3lEthyH3p9v4RX7IYCe4lD3y9kb-Jh95FewQfbGXvd_cnlE95r4nGJ14bPJ_QDWMiXDUwrGWVgfJW5Y3SqiAbF2kpS0udgBZtDvvlE0XoNSuTGitFjlz9I9gGMnVkydMdJJ-McxB3nERmcjASKb5DYfw6kKxLYpXVUVOlS6kTuvqriiJDJILxPi8jfZYbj4lfT6PxAa0UiHQSCQaq_9azI0nyKQvnJYCzbtyuV59cnerlacRpKUzOt3NOIKUrFS8QgxVC-YfpXnTld8INVzEQX6IaloCs9LnuSjIKysWwjkq2rxJycHny4Epnp__BRCCRwLACQAA

<details> <summary>Prefer JSON? The same config, expanded</summary>

Save this to data/configs/<name>.json, or import it via the Load Configuration dialog.

{
  "_metadata": {
    "name": "CHUD-Qwen3.6-27B",
    "description": "Training configuration shared from a Merlina-trained model.",
    "tags": [],
    "schema": "merlina/training-config",
    "schema_version": 1,
    "merlina_version": "2.1.1"
  },
  "base_model": "nbeerbower/BigBubba-Qwen3.6-27B",
  "output_name": "CHUD-Qwen3.6-27B",
  "use_lora": true,
  "lora_r": 32,
  "lora_alpha": 64,
  "lora_dropout": 0.0,
  "target_modules": [
    "up_proj",
    "down_proj",
    "gate_proj",
    "k_proj",
    "q_proj",
    "v_proj",
    "o_proj"
  ],
  "modules_to_save": [],
  "lora_task_type": "CAUSAL_LM",
  "learning_rate": 8e-06,
  "num_epochs": 2,
  "batch_size": 2,
  "gradient_accumulation_steps": 4,
  "max_length": 2048,
  "max_prompt_length": 1536,
  "model_type": "auto",
  "training_mode": "orpo",
  "beta": 0.1,
  "label_smoothing": 0.0,
  "gamma": 0.5,
  "image_resolution": 1024,
  "lora_rank": 32,
  "lora_use_dora": false,
  "mid_training_samples": true,
  "dataset": {
    "source": {
      "source_type": "huggingface",
      "repo_id": "nbeerbower/weasel-dpo",
      "split": "train",
      "streaming": false,
      "streaming_batch_size": 10000
    },
    "additional_sources": [
      {
        "source_type": "huggingface",
        "repo_id": "nbeerbower/seX-ai-dpo",
        "split": "train",
        "streaming": false,
        "streaming_batch_size": 10000,
        "column_mapping": {
          "prompt": "prompt",
          "chosen": "chosen",
          "rejected": "rejected"
        }
      },
      {
        "source_type": "huggingface",
        "repo_id": "nbeerbower/grok-politically-incorrect-dpo",
        "split": "train",
        "streaming": false,
        "streaming_batch_size": 10000,
        "column_mapping": {
          "prompt": "prompt",
          "chosen": "chosen",
          "rejected": "rejected"
        }
      }
    ],
    "format": {
      "format_type": "tokenizer",
      "enable_thinking": true,
      "auto_detect_thinking": true
    },
    "model_name": "nbeerbower/BigBubba-Qwen3.6-27B",
    "column_mapping": {
      "prompt": "prompt",
      "chosen": "chosen",
      "rejected": "rejected"
    },
    "convert_messages_format": true,
    "deduplicate": false,
    "dedupe_strategy": "prompt_chosen",
    "test_size": 0.01,
    "system_prompt_mode": "fill_empty",
    "training_mode": "orpo"
  },
  "seed": 42,
  "max_grad_norm": 1.0,
  "warmup_ratio": 0.1,
  "eval_steps": 0.2,
  "use_4bit": false,
  "use_wandb": true,
  "push_to_hub": true,
  "merge_lora_before_upload": true,
  "hf_hub_private": true,
  "export_gguf": false,
  "gguf_quant_types": [
    "Q4_K_M"
  ],
  "keep_gguf_fp16": false,
  "shuffle_dataset": true,
  "weight_decay": 0.01,
  "lr_scheduler_type": "cosine",
  "gradient_checkpointing": true,
  "logging_steps": 1,
  "optimizer_type": "paged_adamw_8bit",
  "adam_beta1": 0.9,
  "adam_beta2": 0.999,
  "adam_epsilon": 1e-08,
  "adafactor_relative_step": false,
  "adafactor_scale_parameter": false,
  "adafactor_warmup_init": false,
  "adafactor_decay_rate": -0.8,
  "adafactor_clip_threshold": 1.0,
  "attn_implementation": "auto",
  "use_liger": false,
  "torch_compile": false,
  "eval_on_start": false,
  "multi_gpu_strategy": "auto"
}

</details>

---

Merlina on GitHub