ChuckMcSneed/dolphin-2.9.1-dbrx-llamacppfixed

GGUF fix

This is the same model as cognitivecomputations/dolphin-2.9.1-dbrx, but with gguf fixes made by Kenjoyer applied(thanks a lot!). This model can be converted into gguf using llama.cpp.

Benchmarks and personal opinion

NeoEvalPlusN_benchmark

My meme benchmark.

Name	Quant	Size	B	C	D	S	P	total	BCD	SP
cognitivecomputations/dolphin-2.9.1-dbrx	Q6_K	16x12B	3	1	3	4	6	17	7	10
cognitivecomputations/dolphin-2.9.1-qwen-110b	Q6_K	110B	0	1	3	3.75	4.25	12	4	8
databricks/dbrx-instruct	Q6_K	16x12B	0	0	0	6.5	4.5	11	0	11
cognitivecomputations/dolphin-2.2-70b	Q6_K	70B	0	1	1	4.5	4.5	11	2	9
Maximum	n/a	n/a	3	2	3	8	6	22	8	14

More compliant than the official instruct tune(BCD). To my surprise, performed much better overall than qwen-110b tuned on the same dataset. Wrote 6 perfect poems(P column), which is very unusual. Only models from goliath family and more recent llama-3-70b-instruct could do that. Stylized writing tests(S column) were a bit disappointing, Dolphin is not famous for that. In practical use, did perform better than the official tune. Still knows a lot, just like the official tune. Writing is not great, wouldn't use it over Command-r+, unless I need to know some obscure facts. Feels like quantization hurts it a lot more than dense models.

Verdict: Meh, just like the other dolphins. Eric, no disrespect, but you need to get better datasets. GPTslop really hurts practical performance of the model.

Original model card below

Dolphin 2.9.1 DBRX 🐬

Curated and trained by Eric Hartford, Lucas Atkins, and Fernando Fernandes, and Cognitive Computations

Discord: https://discord.gg/cognitivecomputations

Our appreciation for the sponsors of Dolphin 2.9.1:

• Crusoe Cloud - provided excellent on-demand 8xH100 node

This model is based on databricks/dbrx-base, and is governed by databricks-open-model-license

The base model has 32k context, and the full-weight fine-tuning was with 4k sequence length.

This model was trained FFT on parameters selected by Laser Scanner, using ChatML prompt template format.

example:

<|im_start|>system
You are Dolphin, a helpful AI assistant.<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

Dolphin-2.9.1 has a variety of instruction, conversational, and coding skills. It also has initial agentic abilities and supports function calling.

Dolphin is uncensored. We have filtered the dataset to remove alignment and bias. This makes the model more compliant. You are advised to implement your own alignment layer before exposing the model as a service. It will be highly compliant with any requests, even unethical ones. Please read my blog post about uncensored models. https://erichartford.com/uncensored-models You are responsible for any content you create using this model. Enjoy responsibly.

Dolphin is licensed according to Meta's Llama license. We grant permission for any use, including commercial, that falls within accordance with Meta's Llama-3 license. Dolphin was trained on data generated from GPT4, among other models.

Evals

Training

See axolotl config

axolotl version: 0.4.0

base_model: /workspace/axolotl/dbrx-checkpoint
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
trust_remote_code: true

load_in_8bit: false
# load_in_4bit: true
strict: false

# adapter: qlora
# lora_modules_to_save: [embed_tokens, lm_head]

# lora_r: 32
# lora_alpha: 16
# lora_dropout: 0.05
# lora_target_linear: false
# lora_fan_in_fan_out:

datasets:
  - path: /workspace/datasets/dolphin-2.9/dolphin201-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  # - path: /workspace/datasets/dolphin-2.9/Ultrachat200kunfiltered.jsonl
  #   type: sharegpt
  #   conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/dolphin-coder-translate-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/dolphin-coder-codegen-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/m-a-p_Code-Feedback-sharegpt-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/m-a-p_CodeFeedback-Filtered-Instruction-sharegpt-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/not_samantha_norefusals.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/Orca-Math-resort-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/agent_instruct_react_unfiltered.jsonl
    type: sharegpt  
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/toolbench_instruct_j1s1_3k_unfiltered.jsonl
    type: sharegpt  
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/toolbench_negative_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/toolbench_react_10p_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/toolbench_tflan_cot_30p_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/openhermes200k_unfiltered.jsonl
    type: sharegpt 
    conversation: chatml
  # - path: /workspace/datasets/dolphin-2.9/SystemConversations.jsonl
  #   type: sharegpt
  #   conversation: chatml

chat_template: chatml

unfrozen_parameters:
- ^lm_head.weight$
# ffn.experts.mlp_experts.0.v1 layers
- transformer.blocks.30.ffn.experts.mlp_experts.0.v1
- transformer.blocks.32.ffn.experts.mlp_experts.0.v1
- transformer.blocks.25.ffn.experts.mlp_experts.0.v1
- transformer.blocks.15.ffn.experts.mlp_experts.0.v1
- transformer.blocks.22.ffn.experts.mlp_experts.0.v1
- transformer.blocks.31.ffn.experts.mlp_experts.0.v1
- transformer.blocks.7.ffn.experts.mlp_experts.0.v1
- transformer.blocks.21.ffn.experts.mlp_experts.0.v1
- transformer.blocks.8.ffn.experts.mlp_experts.0.v1
- transformer.blocks.23.ffn.experts.mlp_experts.0.v1
# ffn.experts.mlp_experts.0.w1 layers
- transformer.blocks.7.ffn.experts.mlp_experts.0.w1
- transformer.blocks.8.ffn.experts.mlp_experts.0.w1
- transformer.blocks.30.ffn.experts.mlp_experts.0.w1
- transformer.blocks.4.ffn.experts.mlp_experts.0.w1
- transformer.blocks.0.ffn.experts.mlp_experts.0.w1
- transformer.blocks.32.ffn.experts.mlp_experts.0.w1
- transformer.blocks.6.ffn.experts.mlp_experts.0.w1
- transformer.blocks.3.ffn.experts.mlp_experts.0.w1
- transformer.blocks.25.ffn.experts.mlp_experts.0.w1
- transformer.blocks.5.ffn.experts.mlp_experts.0.w1
# ffn.experts.mlp_experts.0.w2 layers
- transformer.blocks.25.ffn.experts.mlp_experts.0.w2
- transformer.blocks.22.ffn.experts.mlp_experts.0.w2
- transformer.blocks.27.ffn.experts.mlp_experts.0.w2
- transformer.blocks.26.ffn.experts.mlp_experts.0.w2
- transformer.blocks.4.ffn.experts.mlp_experts.0.w2
- transformer.blocks.29.ffn.experts.mlp_experts.0.w2
- transformer.blocks.32.ffn.experts.mlp_experts.0.w2
- transformer.blocks.5.ffn.experts.mlp_experts.0.w2
- transformer.blocks.7.ffn.experts.mlp_experts.0.w2
- transformer.blocks.3.ffn.experts.mlp_experts.0.w2
# ffn.experts.mlp_experts.1.v1 layers
- transformer.blocks.27.ffn.experts.mlp_experts.1.v1
- transformer.blocks.25.ffn.experts.mlp_experts.1.v1
- transformer.blocks.29.ffn.experts.mlp_experts.1.v1
- transformer.blocks.33.ffn.experts.mlp_experts.1.v1
- transformer.blocks.23.ffn.experts.mlp_experts.1.v1
- transformer.blocks.30.ffn.experts.mlp_experts.1.v1
- transformer.blocks.6.ffn.experts.mlp_experts.1.v1
- transformer.blocks.21.ffn.experts.mlp_experts.1.v1
- transformer.blocks.15.ffn.experts.mlp_experts.1.v1
- transformer.blocks.7.ffn.experts.mlp_experts.1.v1
# ffn.experts.mlp_experts.1.w1 layers
- transformer.blocks.0.ffn.experts.mlp_experts.1.w1
- transformer.blocks.6.ffn.experts.mlp_experts.1.w1
- transformer.blocks.7.ffn.experts.mlp_experts.1.w1
- transformer.blocks.4.ffn.experts.mlp_experts.1.w1
- transformer.blocks.8.ffn.experts.mlp_experts.1.w1
- transformer.blocks.29.ffn.experts.mlp_experts.1.w1
- transformer.blocks.33.ffn.experts.mlp_experts.1.w1
- transformer.blocks.27.ffn.experts.mlp_experts.1.w1
- transformer.blocks.1.ffn.experts.mlp_experts.1.w1
- transformer.blocks.10.ffn.experts.mlp_experts.1.w1
# ffn.experts.mlp_experts.1.w2 layers
- transformer.blocks.25.ffn.experts.mlp_experts.1.w2
- transformer.blocks.23.ffn.experts.mlp_experts.1.w2
- transformer.blocks.27.ffn.experts.mlp_experts.1.w2
- transformer.blocks.29.ffn.experts.mlp_experts.1.w2
- transformer.blocks.31.ffn.experts.mlp_experts.1.w2
- transformer.blocks.4.ffn.experts.mlp_experts.1.w2
- transformer.blocks.32.ffn.experts.mlp_experts.1.w2
- transformer.blocks.30.ffn.experts.mlp_experts.1.w2
- transformer.blocks.21.ffn.experts.mlp_experts.1.w2
- transformer.blocks.33.ffn.experts.mlp_experts.1.w2
# ffn.experts.mlp_experts.10.v1 layers
- transformer.blocks.28.ffn.experts.mlp_experts.10.v1
- transformer.blocks.34.ffn.experts.mlp_experts.10.v1
- transformer.blocks.33.ffn.experts.mlp_experts.10.v1
- transformer.blocks.26.ffn.experts.mlp_experts.10.v1
- transformer.blocks.32.ffn.experts.mlp_experts.10.v1
- transformer.blocks.30.ffn.experts.mlp_experts.10.v1
- transformer.blocks.36.ffn.experts.mlp_experts.10.v1
- transformer.blocks.24.ffn.experts.mlp_experts.10.v1
- transformer.blocks.20.ffn.experts.mlp_experts.10.v1
- transformer.blocks.35.ffn.experts.mlp_experts.10.v1
# ffn.experts.mlp_experts.10.w1 layers
- transformer.blocks.24.ffn.experts.mlp_experts.10.w1
- transformer.blocks.33.ffn.experts.mlp_experts.10.w1
- transformer.blocks.8.ffn.experts.mlp_experts.10.w1
- transformer.blocks.7.ffn.experts.mlp_experts.10.w1
- transformer.blocks.34.ffn.experts.mlp_experts.10.w1
- transformer.blocks.28.ffn.experts.mlp_experts.10.w1
- transformer.blocks.30.ffn.experts.mlp_experts.10.w1
- transformer.blocks.1.ffn.experts.mlp_experts.10.w1
- transformer.blocks.3.ffn.experts.mlp_experts.10.w1
- transformer.blocks.5.ffn.experts.mlp_experts.10.w1
# ffn.experts.mlp_experts.10.w2 layers
- transformer.blocks.24.ffn.experts.mlp_experts.10.w2
- transformer.blocks.28.ffn.experts.mlp_experts.10.w2
- transformer.blocks.23.ffn.experts.mlp_experts.10.w2
- transformer.blocks.30.ffn.experts.mlp_experts.10.w2
- transformer.blocks.32.ffn.experts.mlp_experts.10.w2
- transformer.blocks.3.ffn.experts.mlp_experts.10.w2
- transformer.blocks.33.ffn.experts.mlp_experts.10.w2
- transformer.blocks.26.ffn.experts.mlp_experts.10.w2
- transformer.blocks.2.ffn.experts.mlp_experts.10.w2
- transformer.blocks.20.ffn.experts.mlp_experts.10.w2
# ffn.experts.mlp_experts.11.w1 layers
- transformer.blocks.6.ffn.experts.mlp_experts.11.w1
- transformer.blocks.8.ffn.experts.mlp_experts.11.w1
- transformer.blocks.9.ffn.experts.mlp_experts.11.w1
- transformer.blocks.0.ffn.experts.mlp_experts.11.w1
- transformer.blocks.10.ffn.experts.mlp_experts.11.w1
- transformer.blocks.28.ffn.experts.mlp_experts.11.w1
- transformer.blocks.3.ffn.experts.mlp_experts.11.w1
- transformer.blocks.5.ffn.experts.mlp_experts.11.w1
- transformer.blocks.33.ffn.experts.mlp_experts.11.w1
- transformer.blocks.13.ffn.experts.mlp_experts.11.w1
# ffn.experts.mlp_experts.11.w2 layers
- transformer.blocks.27.ffn.experts.mlp_experts.11.w2
- transformer.blocks.24.ffn.experts.mlp_experts.11.w2
- transformer.blocks.29.ffn.experts.mlp_experts.11.w2
- transformer.blocks.30.ffn.experts.mlp_experts.11.w2
- transformer.blocks.22.ffn.experts.mlp_experts.11.w2
- transformer.blocks.6.ffn.experts.mlp_experts.11.w2
- transformer.blocks.25.ffn.experts.mlp_experts.11.w2
- transformer.blocks.7.ffn.experts.mlp_experts.11.w2
- transformer.blocks.28.ffn.experts.mlp_experts.11.w2
- transformer.blocks.5.ffn.experts.mlp_experts.11.w2
# ffn.experts.mlp_experts.12.v1 layers
- transformer.blocks.30.ffn.experts.mlp_experts.12.v1
- transformer.blocks.21.ffn.experts.mlp_experts.12.v1
- transformer.blocks.27.ffn.experts.mlp_experts.12.v1
- transformer.blocks.28.ffn.experts.mlp_experts.12.v1
- transformer.blocks.29.ffn.experts.mlp_experts.12.v1
- transformer.blocks.8.ffn.experts.mlp_experts.12.v1
- transformer.blocks.10.ffn.experts.mlp_experts.12.v1
- transformer.blocks.23.ffn.experts.mlp_experts.12.v1
- transformer.blocks.6.ffn.experts.mlp_experts.12.v1
- transformer.blocks.20.ffn.experts.mlp_experts.12.v1
# ffn.experts.mlp_experts.12.w1 layers
- transformer.blocks.8.ffn.experts.mlp_experts.12.w1
- transformer.blocks.1.ffn.experts.mlp_experts.12.w1
- transformer.blocks.0.ffn.experts.mlp_experts.12.w1
- transformer.blocks.6.ffn.experts.mlp_experts.12.w1
- transformer.blocks.9.ffn.experts.mlp_experts.12.w1
- transformer.blocks.2.ffn.experts.mlp_experts.12.w1
- transformer.blocks.10.ffn.experts.mlp_experts.12.w1
- transformer.blocks.17.ffn.experts.mlp_experts.12.w1
- transformer.blocks.29.ffn.experts.mlp_experts.12.w1
- transformer.blocks.21.ffn.experts.mlp_experts.12.w1
# ffn.experts.mlp_experts.12.w2 layers
- transformer.blocks.6.ffn.experts.mlp_experts.12.w2
- transformer.blocks.25.ffn.experts.mlp_experts.12.w2
- transformer.blocks.27.ffn.experts.mlp_experts.12.w2
- transformer.blocks.8.ffn.experts.mlp_experts.12.w2
- transformer.blocks.31.ffn.experts.mlp_experts.12.w2
- transformer.blocks.21.ffn.experts.mlp_experts.12.w2
- transformer.blocks.2.ffn.experts.mlp_experts.12.w2
- transformer.blocks.29.ffn.experts.mlp_experts.12.w2
- transformer.blocks.32.ffn.experts.mlp_experts.12.w2
- transformer.blocks.30.ffn.experts.mlp_experts.12.w2
# ffn.experts.mlp_experts.13.v1 layers
- transformer.blocks.31.ffn.experts.mlp_experts.13.v1
- transformer.blocks.24.ffn.experts.mlp_experts.13.v1
- transformer.blocks.30.ffn.experts.mlp_experts.13.v1
- transformer.blocks.29.ffn.experts.mlp_experts.13.v1
- transformer.blocks.8.ffn.experts.mlp_experts.13.v1
- transformer.blocks.10.ffn.experts.mlp_experts.13.v1
- transformer.blocks.11.ffn.experts.mlp_experts.13.v1
- transformer.blocks.27.ffn.experts.mlp_experts.13.v1
- transformer.blocks.25.ffn.experts.mlp_experts.13.v1
- transformer.blocks.36.ffn.experts.mlp_experts.13.v1
# ffn.experts.mlp_experts.13.w1 layers
- transformer.blocks.4.ffn.experts.mlp_experts.13.w1
- transformer.blocks.10.ffn.experts.mlp_experts.13.w1
- transformer.blocks.6.ffn.experts.mlp_experts.13.w1
- transformer.blocks.0.ffn.experts.mlp_experts.13.w1
- transformer.blocks.3.ffn.experts.mlp_experts.13.w1
- transformer.blocks.24.ffn.experts.mlp_experts.13.w1
- transformer.blocks.8.ffn.experts.mlp_experts.13.w1
- transformer.blocks.1.ffn.experts.mlp_experts.13.w1
- transformer.blocks.30.ffn.experts.mlp_experts.13.w1
- transformer.blocks.11.ffn.experts.mlp_experts.13.w1
# ffn.experts.mlp_experts.13.w2 layers
- transformer.blocks.24.ffn.experts.mlp_experts.13.w2
- transformer.blocks.20.ffn.experts.mlp_experts.13.w2
- transformer.blocks.25.ffn.experts.mlp_experts.13.w2
- transformer.blocks.27.ffn.experts.mlp_experts.13.w2
- transformer.blocks.3.ffn.experts.mlp_experts.13.w2
- transformer.blocks.4.ffn.experts.mlp_experts.13.w2
- transformer.blocks.29.ffn.experts.mlp_experts.13.w2
- transformer.blocks.6.ffn.experts.mlp_experts.13.w2
- transformer.blocks.30.ffn.experts.mlp_experts.13.w2
- transformer.blocks.31.ffn.experts.mlp_experts.13.w2
# ffn.experts.mlp_experts.14.v1 layers
- transformer.blocks.28.ffn.experts.mlp_experts.14.v1
- transformer.blocks.26.ffn.experts.mlp_experts.14.v1
- transformer.blocks.29.ffn.experts.mlp_experts.14.v1
- transformer.blocks.35.ffn.experts.mlp_experts.14.v1
- transformer.blocks.24.ffn.experts.mlp_experts.14.v1
- transformer.blocks.8.ffn.experts.mlp_experts.14.v1
- transformer.blocks.32.ffn.experts.mlp_experts.14.v1
- transformer.blocks.15.ffn.experts.mlp_experts.14.v1
- transformer.blocks.11.ffn.experts.mlp_experts.14.v1
- transformer.blocks.22.ffn.experts.mlp_experts.14.v1
# ffn.experts.mlp_experts.14.w1 layers
- transformer.blocks.8.ffn.experts.mlp_experts.14.w1
- transformer.blocks.4.ffn.experts.mlp_experts.14.w1
- transformer.blocks.5.ffn.experts.mlp_experts.14.w1
- transformer.blocks.7.ffn.experts.mlp_experts.14.w1
- transformer.blocks.3.ffn.experts.mlp_experts.14.w1
- transformer.blocks.13.ffn.experts.mlp_experts.14.w1
- transformer.blocks.29.ffn.experts.mlp_experts.14.w1
- transformer.blocks.6.ffn.experts.mlp_experts.14.w1
- transformer.blocks.28.ffn.experts.mlp_experts.14.w1
- transformer.blocks.9.ffn.experts.mlp_experts.14.w1
# ffn.experts.mlp_experts.14.w2 layers
- transformer.blocks.26.ffn.experts.mlp_experts.14.w2
- transformer.blocks.24.ffn.experts.mlp_experts.14.w2
- transformer.blocks.29.ffn.experts.mlp_experts.14.w2
- transformer.blocks.28.ffn.experts.mlp_experts.14.w2
- transformer.blocks.31.ffn.experts.mlp_experts.14.w2
- transformer.blocks.5.ffn.experts.mlp_experts.14.w2
- transformer.blocks.4.ffn.experts.mlp_experts.14.w2
- transformer.blocks.32.ffn.experts.mlp_experts.14.w2
- transformer.blocks.6.ffn.experts.mlp_experts.14.w2
- transformer.blocks.22.ffn.experts.mlp_experts.14.w2
# ffn.experts.mlp_experts.15.v1 layers
- transformer.blocks.33.ffn.experts.mlp_experts.15.v1
- transformer.blocks.26.ffn.experts.mlp_experts.15.v1
- transformer.blocks.31.ffn.experts.mlp_experts.15.v1
- transformer.blocks.28.ffn.experts.mlp_experts.15.v1
- transformer.blocks.9.ffn.experts.mlp_experts.15.v1
- transformer.blocks.34.ffn.experts.mlp_experts.15.v1
- transformer.blocks.29.ffn.experts.mlp_experts.15.v1
- transformer.blocks.7.ffn.experts.mlp_experts.15.v1
- transformer.blocks.17.ffn.experts.mlp_experts.15.v1
- transformer.blocks.15.ffn.experts.mlp_experts.15.v1
# ffn.experts.mlp_experts.15.w1 layers
- transformer.blocks.6.ffn.experts.mlp_experts.15.w1
- transformer.blocks.9.ffn.experts.mlp_experts.15.w1
- transformer.blocks.0.ffn.experts.mlp_experts.15.w1
- transformer.blocks.7.ffn.experts.mlp_experts.15.w1
- transformer.blocks.14.ffn.experts.mlp_experts.15.w1
- transformer.blocks.33.ffn.experts.mlp_experts.15.w1
- transformer.blocks.34.ffn.experts.mlp_experts.15.w1
- transformer.blocks.10.ffn.experts.mlp_experts.15.w1
- transformer.blocks.5.ffn.experts.mlp_experts.15.w1
- transformer.blocks.29.ffn.experts.mlp_experts.15.w1
# ffn.experts.mlp_experts.15.w2 layers
- transformer.blocks.28.ffn.experts.mlp_experts.15.w2
- transformer.blocks.26.ffn.experts.mlp_experts.15.w2
- transformer.blocks.27.ffn.experts.mlp_experts.15.w2
- transformer.blocks.29.ffn.experts.mlp_experts.15.w2
- transformer.blocks.6.ffn.experts.mlp_experts.15.w2
- transformer.blocks.31.ffn.experts.mlp_experts.15.w2
- transformer.blocks.7.ffn.experts.mlp_experts.15.w2
- transformer.blocks.33.ffn.experts.mlp_experts.15.w2
- transformer.blocks.32.ffn.experts.mlp_experts.15.w2
- transformer.blocks.25.ffn.experts.mlp_experts.15.w2
# ffn.experts.mlp_experts.2.v1 layers
- transformer.blocks.31.ffn.experts.mlp_experts.2.v1
- transformer.blocks.27.ffn.experts.mlp_experts.2.v1
- transformer.blocks.28.ffn.experts.mlp_experts.2.v1
- transformer.blocks.30.ffn.experts.mlp_experts.2.v1
- transformer.blocks.23.ffn.experts.mlp_experts.2.v1
- transformer.blocks.32.ffn.experts.mlp_experts.2.v1
- transformer.blocks.35.ffn.experts.mlp_experts.2.v1
- transformer.blocks.7.ffn.experts.mlp_experts.2.v1
- transformer.blocks.21.ffn.experts.mlp_experts.2.v1
- transformer.blocks.15.ffn.experts.mlp_experts.2.v1
# ffn.experts.mlp_experts.2.w1 layers
- transformer.blocks.7.ffn.experts.mlp_experts.2.w1
- transformer.blocks.6.ffn.experts.mlp_experts.2.w1
- transformer.blocks.1.ffn.experts.mlp_experts.2.w1
- transformer.blocks.4.ffn.experts.mlp_experts.2.w1
- transformer.blocks.5.ffn.experts.mlp_experts.2.w1
- transformer.blocks.29.ffn.experts.mlp_experts.2.w1
- transformer.blocks.0.ffn.experts.mlp_experts.2.w1
- transformer.blocks.9.ffn.experts.mlp_experts.2.w1
- transformer.blocks.31.ffn.experts.mlp_experts.2.w1
- transformer.blocks.30.ffn.experts.mlp_experts.2.w1
# ffn.experts.mlp_experts.2.w2 layers
- transformer.blocks.26.ffn.experts.mlp_experts.2.w2
- transformer.blocks.27.ffn.experts.mlp_experts.2.w2
- transformer.blocks.33.ffn.experts.mlp_experts.2.w2
- transformer.blocks.5.ffn.experts.mlp_experts.2.w2
- transformer.blocks.23.ffn.experts.mlp_experts.2.w2
- transformer.blocks.32.ffn.experts.mlp_experts.2.w2
- transformer.blocks.28.ffn.experts.mlp_experts.2.w2
- transformer.blocks.4.ffn.experts.mlp_experts.2.w2
- transformer.blocks.29.ffn.experts.mlp_experts.2.w2
- transformer.blocks.30.ffn.experts.mlp_experts.2.w2
# ffn.experts.mlp_experts.3.v1 layers
- transformer.blocks.28.ffn.experts.mlp_experts.3.v1
- transformer.blocks.33.ffn.experts.mlp_experts.3.v1
- transformer.blocks.36.ffn.experts.mlp_experts.3.v1
- transformer.blocks.29.ffn.experts.mlp_experts.3.v1
- transformer.blocks.30.ffn.experts.mlp_experts.3.v1
- transformer.blocks.7.ffn.experts.mlp_experts.3.v1
- transformer.blocks.14.ffn.experts.mlp_experts.3.v1
- transformer.blocks.10.ffn.experts.mlp_experts.3.v1
- transformer.blocks.31.ffn.experts.mlp_experts.3.v1
- transformer.blocks.21.ffn.experts.mlp_experts.3.v1
# ffn.experts.mlp_experts.3.w1 layers
- transformer.blocks.7.ffn.experts.mlp_experts.3.w1
- transformer.blocks.0.ffn.experts.mlp_experts.3.w1
- transformer.blocks.10.ffn.experts.mlp_experts.3.w1
- transformer.blocks.9.ffn.experts.mlp_experts.3.w1
- transformer.blocks.29.ffn.experts.mlp_experts.3.w1
- transformer.blocks.5.ffn.experts.mlp_experts.3.w1
- transformer.blocks.30.ffn.experts.mlp_experts.3.w1
- transformer.blocks.4.ffn.experts.mlp_experts.3.w1
- transformer.blocks.33.ffn.experts.mlp_experts.3.w1
- transformer.blocks.1.ffn.experts.mlp_experts.3.w1
# ffn.experts.mlp_experts.3.w2 layers
- transformer.blocks.28.ffn.experts.mlp_experts.3.w2
- transformer.blocks.5.ffn.experts.mlp_experts.3.w2
- transformer.blocks.24.ffn.experts.mlp_experts.3.w2
- transformer.blocks.31.ffn.experts.mlp_experts.3.w2
- transformer.blocks.30.ffn.experts.mlp_experts.3.w2
- transformer.blocks.21.ffn.experts.mlp_experts.3.w2
- transformer.blocks.32.ffn.experts.mlp_experts.3.w2
- transformer.blocks.29.ffn.experts.mlp_experts.3.w2
- transformer.blocks.26.ffn.experts.mlp_experts.3.w2
- transformer.blocks.2.ffn.experts.mlp_experts.3.w2
# ffn.experts.mlp_experts.4.v1 layers
- transformer.blocks.34.ffn.experts.mlp_experts.4.v1
- transformer.blocks.31.ffn.experts.mlp_experts.4.v1
- transformer.blocks.26.ffn.experts.mlp_experts.4.v1
- transformer.blocks.24.ffn.experts.mlp_experts.4.v1
- transformer.blocks.14.ffn.experts.mlp_experts.4.v1
- transformer.blocks.32.ffn.experts.mlp_experts.4.v1
- transformer.blocks.7.ffn.experts.mlp_experts.4.v1
- transformer.blocks.6.ffn.experts.mlp_experts.4.v1
- transformer.blocks.20.ffn.experts.mlp_experts.4.v1
- transformer.blocks.9.ffn.experts.mlp_experts.4.v1
# ffn.experts.mlp_experts.4.w1 layers
- transformer.blocks.6.ffn.experts.mlp_experts.4.w1
- transformer.blocks.4.ffn.experts.mlp_experts.4.w1
- transformer.blocks.7.ffn.experts.mlp_experts.4.w1
- transformer.blocks.9.ffn.experts.mlp_experts.4.w1
- transformer.blocks.0.ffn.experts.mlp_experts.4.w1
- transformer.blocks.5.ffn.experts.mlp_experts.4.w1
- transformer.blocks.14.ffn.experts.mlp_experts.4.w1
- transformer.blocks.34.ffn.experts.mlp_experts.4.w1
- transformer.blocks.8.ffn.experts.mlp_experts.4.w1
- transformer.blocks.29.ffn.experts.mlp_experts.4.w1
# ffn.experts.mlp_experts.4.w2 layers
- transformer.blocks.25.ffn.experts.mlp_experts.4.w2
- transformer.blocks.24.ffn.experts.mlp_experts.4.w2
- transformer.blocks.26.ffn.experts.mlp_experts.4.w2
- transformer.blocks.5.ffn.experts.mlp_experts.4.w2
- transformer.blocks.6.ffn.experts.mlp_experts.4.w2
- transformer.blocks.32.ffn.experts.mlp_experts.4.w2
- transformer.blocks.4.ffn.experts.mlp_experts.4.w2
- transformer.blocks.36.ffn.experts.mlp_experts.4.w2
- transformer.blocks.29.ffn.experts.mlp_experts.4.w2
- transformer.blocks.27.ffn.experts.mlp_experts.4.w2
# ffn.experts.mlp_experts.5.v1 layers
- transformer.blocks.35.ffn.experts.mlp_experts.5.v1
- transformer.blocks.30.ffn.experts.mlp_experts.5.v1
- transformer.blocks.28.ffn.experts.mlp_experts.5.v1
- transformer.blocks.32.ffn.experts.mlp_experts.5.v1
- transformer.blocks.27.ffn.experts.mlp_experts.5.v1
- transformer.blocks.26.ffn.experts.mlp_experts.5.v1
- transformer.blocks.33.ffn.experts.mlp_experts.5.v1
- transformer.blocks.29.ffn.experts.mlp_experts.5.v1
- transformer.blocks.8.ffn.experts.mlp_experts.5.v1
- transformer.blocks.7.ffn.experts.mlp_experts.5.v1
# ffn.experts.mlp_experts.5.w1 layers
- transformer.blocks.0.ffn.experts.mlp_experts.5.w1
- transformer.blocks.6.ffn.experts.mlp_experts.5.w1
- transformer.blocks.7.ffn.experts.mlp_experts.5.w1
- transformer.blocks.9.ffn.experts.mlp_experts.5.w1
- transformer.blocks.8.ffn.experts.mlp_experts.5.w1
- transformer.blocks.12.ffn.experts.mlp_experts.5.w1
- transformer.blocks.3.ffn.experts.mlp_experts.5.w1
- transformer.blocks.5.ffn.experts.mlp_experts.5.w1
- transformer.blocks.4.ffn.experts.mlp_experts.5.w1
- transformer.blocks.33.ffn.experts.mlp_experts.5.w1
# ffn.experts.mlp_experts.5.w2 layers
- transformer.blocks.26.ffn.experts.mlp_experts.5.w2
- transformer.blocks.28.ffn.experts.mlp_experts.5.w2
- transformer.blocks.6.ffn.experts.mlp_experts.5.w2
- transformer.blocks.33.ffn.experts.mlp_experts.5.w2
- transformer.blocks.5.ffn.experts.mlp_experts.5.w2
- transformer.blocks.27.ffn.experts.mlp_experts.5.w2
- transformer.blocks.3.ffn.experts.mlp_experts.5.w2
- transformer.blocks.29.ffn.experts.mlp_experts.5.w2
- transformer.blocks.25.ffn.experts.mlp_experts.5.w2
- transformer.blocks.7.ffn.experts.mlp_experts.5.w2
# ffn.experts.mlp_experts.6.v1 layers
- transformer.blocks.34.ffn.experts.mlp_experts.6.v1
- transformer.blocks.31.ffn.experts.mlp_experts.6.v1
- transformer.blocks.30.ffn.experts.mlp_experts.6.v1
- transformer.blocks.26.ffn.experts.mlp_experts.6.v1
- transformer.blocks.35.ffn.experts.mlp_experts.6.v1
- transformer.blocks.20.ffn.experts.mlp_experts.6.v1
- transformer.blocks.15.ffn.experts.mlp_experts.6.v1
- transformer.blocks.29.ffn.experts.mlp_experts.6.v1
- transformer.blocks.10.ffn.experts.mlp_experts.6.v1
- transformer.blocks.24.ffn.experts.mlp_experts.6.v1
# ffn.experts.mlp_experts.6.w1 layers
- transformer.blocks.0.ffn.experts.mlp_experts.6.w1
- transformer.blocks.10.ffn.experts.mlp_experts.6.w1
- transformer.blocks.9.ffn.experts.mlp_experts.6.w1
- transformer.blocks.30.ffn.experts.mlp_experts.6.w1
- transformer.blocks.4.ffn.experts.mlp_experts.6.w1
- transformer.blocks.34.ffn.experts.mlp_experts.6.w1
- transformer.blocks.26.ffn.experts.mlp_experts.6.w1
- transformer.blocks.2.ffn.experts.mlp_experts.6.w1
- transformer.blocks.29.ffn.experts.mlp_experts.6.w1
- transformer.blocks.8.ffn.experts.mlp_experts.6.w1
# ffn.experts.mlp_experts.6.w2 layers
- transformer.blocks.24.ffn.experts.mlp_experts.6.w2
- transformer.blocks.26.ffn.experts.mlp_experts.6.w2
- transformer.blocks.32.ffn.experts.mlp_experts.6.w2
- transformer.blocks.30.ffn.experts.mlp_experts.6.w2
- transformer.blocks.25.ffn.experts.mlp_experts.6.w2
- transformer.blocks.31.ffn.experts.mlp_experts.6.w2
- transformer.blocks.20.ffn.experts.mlp_experts.6.w2
- transformer.blocks.4.ffn.experts.mlp_experts.6.w2
- transformer.blocks.2.ffn.experts.mlp_experts.6.w2
- transformer.blocks.9.ffn.experts.mlp_experts.6.w2
# ffn.experts.mlp_experts.7.v1 layers
- transformer.blocks.27.ffn.experts.mlp_experts.7.v1
- transformer.blocks.28.ffn.experts.mlp_experts.7.v1
- transformer.blocks.33.ffn.experts.mlp_experts.7.v1
- transformer.blocks.29.ffn.experts.mlp_experts.7.v1
- transformer.blocks.24.ffn.experts.mlp_experts.7.v1
- transformer.blocks.11.ffn.experts.mlp_experts.7.v1
- transformer.blocks.12.ffn.experts.mlp_experts.7.v1
- transformer.blocks.10.ffn.experts.mlp_experts.7.v1
- transformer.blocks.23.ffn.experts.mlp_experts.7.v1
- transformer.blocks.34.ffn.experts.mlp_experts.7.v1
# ffn.experts.mlp_experts.7.w1 layers
- transformer.blocks.12.ffn.experts.mlp_experts.7.w1
- transformer.blocks.0.ffn.experts.mlp_experts.7.w1
- transformer.blocks.5.ffn.experts.mlp_experts.7.w1
- transformer.blocks.29.ffn.experts.mlp_experts.7.w1
- transformer.blocks.10.ffn.experts.mlp_experts.7.w1
- transformer.blocks.4.ffn.experts.mlp_experts.7.w1
- transformer.blocks.3.ffn.experts.mlp_experts.7.w1
- transformer.blocks.8.ffn.experts.mlp_experts.7.w1
- transformer.blocks.34.ffn.experts.mlp_experts.7.w1
- transformer.blocks.33.ffn.experts.mlp_experts.7.w1
# ffn.experts.mlp_experts.7.w2 layers
- transformer.blocks.23.ffn.experts.mlp_experts.7.w2
- transformer.blocks.24.ffn.experts.mlp_experts.7.w2
- transformer.blocks.31.ffn.experts.mlp_experts.7.w2
- transformer.blocks.28.ffn.experts.mlp_experts.7.w2
- transformer.blocks.27.ffn.experts.mlp_experts.7.w2
- transformer.blocks.5.ffn.experts.mlp_experts.7.w2
- transformer.blocks.25.ffn.experts.mlp_experts.7.w2
- transformer.blocks.29.ffn.experts.mlp_experts.7.w2
- transformer.blocks.3.ffn.experts.mlp_experts.7.w2
- transformer.blocks.33.ffn.experts.mlp_experts.7.w2
# ffn.experts.mlp_experts.8.v1 layers
- transformer.blocks.30.ffn.experts.mlp_experts.8.v1
- transformer.blocks.27.ffn.experts.mlp_experts.8.v1
- transformer.blocks.20.ffn.experts.mlp_experts.8.v1
- transformer.blocks.32.ffn.experts.mlp_experts.8.v1
- transformer.blocks.34.ffn.experts.mlp_experts.8.v1
- transformer.blocks.33.ffn.experts.mlp_experts.8.v1
- transformer.blocks.9.ffn.experts.mlp_experts.8.v1
- transformer.blocks.7.ffn.experts.mlp_experts.8.v1
- transformer.blocks.6.ffn.experts.mlp_experts.8.v1
- transformer.blocks.24.ffn.experts.mlp_experts.8.v1
# ffn.experts.mlp_experts.8.w1 layers
- transformer.blocks.7.ffn.experts.mlp_experts.8.w1
- transformer.blocks.6.ffn.experts.mlp_experts.8.w1
- transformer.blocks.0.ffn.experts.mlp_experts.8.w1
- transformer.blocks.9.ffn.experts.mlp_experts.8.w1
- transformer.blocks.3.ffn.experts.mlp_experts.8.w1
- transformer.blocks.2.ffn.experts.mlp_experts.8.w1
- transformer.blocks.8.ffn.experts.mlp_experts.8.w1
- transformer.blocks.30.ffn.experts.mlp_experts.8.w1
- transformer.blocks.24.ffn.experts.mlp_experts.8.w1
- transformer.blocks.1.ffn.experts.mlp_experts.8.w1
# ffn.experts.mlp_experts.8.w2 layers
- transformer.blocks.32.ffn.experts.mlp_experts.8.w2
- transformer.blocks.24.ffn.experts.mlp_experts.8.w2
- transformer.blocks.27.ffn.experts.mlp_experts.8.w2
- transformer.blocks.30.ffn.experts.mlp_experts.8.w2
- transformer.blocks.31.ffn.experts.mlp_experts.8.w2
- transformer.blocks.28.ffn.experts.mlp_experts.8.w2
- transformer.blocks.2.ffn.experts.mlp_experts.8.w2
- transformer.blocks.3.ffn.experts.mlp_experts.8.w2
- transformer.blocks.23.ffn.experts.mlp_experts.8.w2
- transformer.blocks.29.ffn.experts.mlp_experts.8.w2
# ffn.experts.mlp_experts.9.v1 layers
- transformer.blocks.31.ffn.experts.mlp_experts.9.v1
- transformer.blocks.27.ffn.experts.mlp_experts.9.v1
- transformer.blocks.29.ffn.experts.mlp_experts.9.v1
- transformer.blocks.33.ffn.experts.mlp_experts.9.v1
- transformer.blocks.25.ffn.experts.mlp_experts.9.v1
- transformer.blocks.14.ffn.experts.mlp_experts.9.v1
- transformer.blocks.32.ffn.experts.mlp_experts.9.v1
- transformer.blocks.7.ffn.experts.mlp_experts.9.v1
- transformer.blocks.9.ffn.experts.mlp_experts.9.v1
- transformer.blocks.34.ffn.experts.mlp_experts.9.v1
# ffn.experts.mlp_experts.9.w1 layers
- transformer.blocks.7.ffn.experts.mlp_experts.9.w1
- transformer.blocks.1.ffn.experts.mlp_experts.9.w1
- transformer.blocks.9.ffn.experts.mlp_experts.9.w1
- transformer.blocks.2.ffn.experts.mlp_experts.9.w1
- transformer.blocks.27.ffn.experts.mlp_experts.9.w1
- transformer.blocks.12.ffn.experts.mlp_experts.9.w1
- transformer.blocks.4.ffn.experts.mlp_experts.9.w1
- transformer.blocks.6.ffn.experts.mlp_experts.9.w1
- transformer.blocks.19.ffn.experts.mlp_experts.9.w1
- transformer.blocks.8.ffn.experts.mlp_experts.9.w1
# ffn.experts.mlp_experts.9.w2 layers
- transformer.blocks.26.ffn.experts.mlp_experts.9.w2
- transformer.blocks.25.ffn.experts.mlp_experts.9.w2
- transformer.blocks.28.ffn.experts.mlp_experts.9.w2
- transformer.blocks.27.ffn.experts.mlp_experts.9.w2
- transformer.blocks.31.ffn.experts.mlp_experts.9.w2
- transformer.blocks.29.ffn.experts.mlp_experts.9.w2
- transformer.blocks.7.ffn.experts.mlp_experts.9.w2
- transformer.blocks.34.ffn.experts.mlp_experts.9.w2
- transformer.blocks.2.ffn.experts.mlp_experts.9.w2
- transformer.blocks.33.ffn.experts.mlp_experts.9.w2
# ffn.router.layer layers
- transformer.blocks.2.ffn.router.layer
- transformer.blocks.3.ffn.router.layer
- transformer.blocks.4.ffn.router.layer
- transformer.blocks.5.ffn.router.layer
- transformer.blocks.6.ffn.router.layer
- transformer.blocks.7.ffn.router.layer
- transformer.blocks.8.ffn.router.layer
- transformer.blocks.9.ffn.router.layer
- transformer.blocks.10.ffn.router.layer
- transformer.blocks.11.ffn.router.layer
# norm_attn_norm.attn.Wqkv layers
- transformer.blocks.16.norm_attn_norm.attn.Wqkv
- transformer.blocks.15.norm_attn_norm.attn.Wqkv
- transformer.blocks.11.norm_attn_norm.attn.Wqkv
- transformer.blocks.14.norm_attn_norm.attn.Wqkv
- transformer.blocks.12.norm_attn_norm.attn.Wqkv
- transformer.blocks.20.norm_attn_norm.attn.Wqkv
- transformer.blocks.10.norm_attn_norm.attn.Wqkv
- transformer.blocks.9.norm_attn_norm.attn.Wqkv
- transformer.blocks.19.norm_attn_norm.attn.Wqkv
- transformer.blocks.18.norm_attn_norm.attn.Wqkv
# norm_attn_norm.attn.out_proj layers
- transformer.blocks.1.norm_attn_norm.attn.out_proj
- transformer.blocks.18.norm_attn_norm.attn.out_proj
- transformer.blocks.2.norm_attn_norm.attn.out_proj
- transformer.blocks.16.norm_attn_norm.attn.out_proj
- transformer.blocks.0.norm_attn_norm.attn.out_proj
- transformer.blocks.39.norm_attn_norm.attn.out_proj
- transformer.blocks.23.norm_attn_norm.attn.out_proj
- transformer.blocks.8.norm_attn_norm.attn.out_proj
- transformer.blocks.24.norm_attn_norm.attn.out_proj
- transformer.blocks.19.norm_attn_norm.attn.out_proj
# norm_attn_norm.norm_1 layers
- transformer.blocks.0.norm_attn_norm.norm_1
- transformer.blocks.1.norm_attn_norm.norm_1
- transformer.blocks.2.norm_attn_norm.norm_1
- transformer.blocks.3.norm_attn_norm.norm_1
- transformer.blocks.4.norm_attn_norm.norm_1
- transformer.blocks.5.norm_attn_norm.norm_1
- transformer.blocks.6.norm_attn_norm.norm_1
- transformer.blocks.7.norm_attn_norm.norm_1
- transformer.blocks.8.norm_attn_norm.norm_1
- transformer.blocks.9.norm_attn_norm.norm_1
# norm_attn_norm.norm_2 layers
- transformer.blocks.0.norm_attn_norm.norm_2
- transformer.blocks.1.norm_attn_norm.norm_2
- transformer.blocks.2.norm_attn_norm.norm_2
- transformer.blocks.3.norm_attn_norm.norm_2
- transformer.blocks.4.norm_attn_norm.norm_2
- transformer.blocks.5.norm_attn_norm.norm_2
- transformer.blocks.6.norm_attn_norm.norm_2
- transformer.blocks.7.norm_attn_norm.norm_2
- transformer.blocks.8.norm_attn_norm.norm_2
- transformer.blocks.9.norm_attn_norm.norm_2
# transformer.norm_f layers
# transformer.wte layers
# ffn.experts.mlp_experts.11.v1 layers
- transformer.blocks.29.ffn.experts.mlp_experts.11.v1
- transformer.blocks.27.ffn.experts.mlp_experts.11.v1
- transformer.blocks.30.ffn.experts.mlp_experts.11.v1
- transformer.blocks.28.ffn.experts.mlp_experts.11.v1
- transformer.blocks.22.ffn.experts.mlp_experts.11.v1
- transformer.blocks.7.ffn.experts.mlp_experts.11.v1
- transformer.blocks.24.ffn.experts.mlp_experts.11.v1
- transformer.blocks.8.ffn.experts.mlp_experts.11.v1
- transformer.blocks.6.ffn.experts.mlp_experts.11.v1
- transformer.blocks.12.ffn.experts.mlp_experts.11.v1



dataset_prepared_path: dbrx2
val_set_size: 0.01
output_dir: ./out

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

wandb_project: dolphin-2.9-Dbrx
wandb_watch:
wandb_run_id:
wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 1
num_epochs: 1
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 1e-5

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: true

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
early_stopping_patience:
# resume_from_checkpoint: /workspace/axolotl/dbrx-checkpoint
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 10
evals_per_epoch: 4
eval_table_size:
saves_per_epoch: 4
save_total_limit: 2
save_steps:
debug:
deepspeed: /workspace/axolotl/deepspeed_configs/zero3_bf16_cpuoffload_params.json
weight_decay: 0.05
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<|endoftext|>"
  eos_token: "<|im_end|>"
  pad_token: "<|pad|>"
  unk_token: "<|endoftext|>"
tokens:
  - "<|im_start|>"
  - "<|im_end|>"
  

out

This model was trained from scratch on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4336

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 8
• total_train_batch_size: 64
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss
0.4009	0.0	1	0.4328
0.413	0.25	587	0.4408
0.3626	0.5	1174	0.4368
0.3896	0.75	1761	0.4336

Framework versions

• Transformers 4.40.0.dev0
• Pytorch 2.2.2+cu121
• Datasets 2.15.0
• Tokenizers 0.15.0