Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,431 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: llama3.1
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: llama3.1
|
| 3 |
+
base_model: meta-llama/Meta-Llama-3.1-8B
|
| 4 |
+
tags:
|
| 5 |
+
- generated_from_trainer
|
| 6 |
+
datasets:
|
| 7 |
+
- cognitivecomputations/Dolphin-2.9
|
| 8 |
+
- m-a-p/CodeFeedback-Filtered-Instruction
|
| 9 |
+
- cognitivecomputations/dolphin-coder
|
| 10 |
+
- cognitivecomputations/samantha-data
|
| 11 |
+
- microsoft/orca-math-word-problems-200k
|
| 12 |
+
- mlabonne/FineTome-100k
|
| 13 |
+
- arcee/agent_data
|
| 14 |
+
- PawanKrd/math-gpt-4o-200k
|
| 15 |
+
- cognitivecomputations/SystemChat-2.0
|
| 16 |
+
---
|
| 17 |
+
|
| 18 |
+
|
| 19 |
+
## Description
|
| 20 |
+
This repo contains GGUF format model files for dolphin-2.9.4-llama3.1-8b.
|
| 21 |
+
|
| 22 |
+
## Files Provided
|
| 23 |
+
| Name | Quant | Bits | File Size | Remark |
|
| 24 |
+
| ----------------------------------- | ----- | ---- | --------- | -------------------------------- |
|
| 25 |
+
| dolphin-2.9.4-llama3.1-8b.Q2_K.gguf | Q2_K | 2 | 3.18 GB | 2.96G, +3.5199 ppl @ Llama-3-8B |
|
| 26 |
+
| dolphin-2.9.4-llama3.1-8b.Q3_K.gguf | Q3_K | 3 | 4.02 GB | 3.74G, +0.6569 ppl @ Llama-3-8B |
|
| 27 |
+
| dolphin-2.9.4-llama3.1-8b.Q4_0.gguf | Q4_0 | 4 | 4.66 GB | 4.34G, +0.4685 ppl @ Llama-3-8B |
|
| 28 |
+
| dolphin-2.9.4-llama3.1-8b.Q4_K.gguf | Q4_K | 4 | 4.92 GB | 4.58G, +0.1754 ppl @ Llama-3-8B |
|
| 29 |
+
| dolphin-2.9.4-llama3.1-8b.Q5_K.gguf | Q5_K | 5 | 5.73 GB | 5.33G, +0.0569 ppl @ Llama-3-8B |
|
| 30 |
+
| dolphin-2.9.4-llama3.1-8b.Q6_K.gguf | Q6_K | 6 | 6.60 GB | 6.14G, +0.0217 ppl @ Llama-3-8B |
|
| 31 |
+
| dolphin-2.9.4-llama3.1-8b.Q8_0.gguf | Q8_0 | 8 | 8.54 GB | 7.96G, +0.0026 ppl @ Llama-3-8B |
|
| 32 |
+
|
| 33 |
+
## Parameters
|
| 34 |
+
| path | type | architecture | rope_theta | sliding_win | max_pos_embed |
|
| 35 |
+
| ----------------------------------------------- | ----- | ---------------- | ---------- | ----------- | ------------- |
|
| 36 |
+
| cognitivecomputations/dolphin-2.9.4-llama3.1-8b | llama | LlamaForCausalLM | 500000.0 | null | 131072 |
|
| 37 |
+
|
| 38 |
+
|
| 39 |
+
# Original Model Card
|
| 40 |
+
|
| 41 |
+
|
| 42 |
+
# Dolphin 2.9.4 Llama 3.1 8b 🐬
|
| 43 |
+
|
| 44 |
+
Curated and trained by Eric Hartford and Cognitive Computations
|
| 45 |
+
|
| 46 |
+
[](https://discord.gg/h3K4XGj2RH)
|
| 47 |
+
Discord: https://discord.gg/h3K4XGj2RH
|
| 48 |
+
|
| 49 |
+
<img src="https://cdn-uploads.huggingface.co/production/uploads/63111b2d88942700629f5771/ldkN1J0WIDQwU4vutGYiD.png" width="600" />
|
| 50 |
+
|
| 51 |
+
Our appreciation for the sponsors of Dolphin 2.9.4:
|
| 52 |
+
- [Crusoe Cloud](https://crusoe.ai/) - provided excellent on-demand 8xL40S node
|
| 53 |
+
|
| 54 |
+
This model is based on Meta Llama 3.1 8b, and is governed by the Llama 3.1 license.
|
| 55 |
+
|
| 56 |
+
The base model has 128K context, and our finetuning used 8192 sequence length.
|
| 57 |
+
|
| 58 |
+
Dolphin 2.9.4 uses ChatML prompt template format.
|
| 59 |
+
|
| 60 |
+
example:
|
| 61 |
+
|
| 62 |
+
```
|
| 63 |
+
<|im_start|>system
|
| 64 |
+
You are Dolphin, a helpful AI assistant.<|im_end|>
|
| 65 |
+
<|im_start|>user
|
| 66 |
+
{prompt}<|im_end|>
|
| 67 |
+
<|im_start|>assistant
|
| 68 |
+
|
| 69 |
+
```
|
| 70 |
+
|
| 71 |
+
Dolphin-2.9.4 has a variety of instruction following, conversational, and coding skills. It also has agentic abilities and supports function calling.
|
| 72 |
+
It is especially trained to obey the system prompt, and follow instructions in many languages.
|
| 73 |
+
|
| 74 |
+
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.
|
| 75 |
+
|
| 76 |
+
|
| 77 |
+
<details><summary>Evals</summary>
|
| 78 |
+
|
| 79 |
+
```
|
| 80 |
+
hf (pretrained=/workspace/axolotl/dolphin-2.9.4-llama3.1-8b-hf,dtype=bfloat16), gen_kwargs: (None), limit: None, num_fewshot: None, batch_size: auto (4)
|
| 81 |
+
| Tasks |Version|Filter|n-shot| Metric | |Value | |Stderr|
|
| 82 |
+
|-----------------------------------------------------------|-------|------|-----:|-----------------------|---|-----:|---|------|
|
| 83 |
+
|leaderboard |N/A |none | 0|acc |↑ |0.2926|± |0.0041|
|
| 84 |
+
| | |none | 0|acc_norm |↑ |0.4513|± |0.0053|
|
| 85 |
+
| | |none | 0|exact_match |↑ |0.0982|± |0.0079|
|
| 86 |
+
| | |none | 0|inst_level_loose_acc |↑ |0.3825|± |N/A |
|
| 87 |
+
| | |none | 0|inst_level_strict_acc |↑ |0.3597|± |N/A |
|
| 88 |
+
| | |none | 0|prompt_level_loose_acc |↑ |0.2421|± |0.0184|
|
| 89 |
+
| | |none | 0|prompt_level_strict_acc|↑ |0.2181|± |0.0178|
|
| 90 |
+
| - leaderboard_bbh |N/A |none | 3|acc_norm |↑ |0.4931|± |0.0061|
|
| 91 |
+
| - leaderboard_bbh_boolean_expressions | 0|none | 3|acc_norm |↑ |0.8000|± |0.0253|
|
| 92 |
+
| - leaderboard_bbh_causal_judgement | 0|none | 3|acc_norm |↑ |0.5615|± |0.0364|
|
| 93 |
+
| - leaderboard_bbh_date_understanding | 0|none | 3|acc_norm |↑ |0.4520|± |0.0315|
|
| 94 |
+
| - leaderboard_bbh_disambiguation_qa | 0|none | 3|acc_norm |↑ |0.6640|± |0.0299|
|
| 95 |
+
| - leaderboard_bbh_formal_fallacies | 0|none | 3|acc_norm |↑ |0.5600|± |0.0315|
|
| 96 |
+
| - leaderboard_bbh_geometric_shapes | 0|none | 3|acc_norm |↑ |0.3640|± |0.0305|
|
| 97 |
+
| - leaderboard_bbh_hyperbaton | 0|none | 3|acc_norm |↑ |0.6320|± |0.0306|
|
| 98 |
+
| - leaderboard_bbh_logical_deduction_five_objects | 0|none | 3|acc_norm |↑ |0.4600|± |0.0316|
|
| 99 |
+
| - leaderboard_bbh_logical_deduction_seven_objects | 0|none | 3|acc_norm |↑ |0.4360|± |0.0314|
|
| 100 |
+
| - leaderboard_bbh_logical_deduction_three_objects | 0|none | 3|acc_norm |↑ |0.6160|± |0.0308|
|
| 101 |
+
| - leaderboard_bbh_movie_recommendation | 0|none | 3|acc_norm |↑ |0.7880|± |0.0259|
|
| 102 |
+
| - leaderboard_bbh_navigate | 0|none | 3|acc_norm |↑ |0.5200|± |0.0317|
|
| 103 |
+
| - leaderboard_bbh_object_counting | 0|none | 3|acc_norm |↑ |0.4520|± |0.0315|
|
| 104 |
+
| - leaderboard_bbh_penguins_in_a_table | 0|none | 3|acc_norm |↑ |0.5205|± |0.0415|
|
| 105 |
+
| - leaderboard_bbh_reasoning_about_colored_objects | 0|none | 3|acc_norm |↑ |0.5120|± |0.0317|
|
| 106 |
+
| - leaderboard_bbh_ruin_names | 0|none | 3|acc_norm |↑ |0.6320|± |0.0306|
|
| 107 |
+
| - leaderboard_bbh_salient_translation_error_detection | 0|none | 3|acc_norm |↑ |0.4320|± |0.0314|
|
| 108 |
+
| - leaderboard_bbh_snarks | 0|none | 3|acc_norm |↑ |0.5843|± |0.0370|
|
| 109 |
+
| - leaderboard_bbh_sports_understanding | 0|none | 3|acc_norm |↑ |0.7040|± |0.0289|
|
| 110 |
+
| - leaderboard_bbh_temporal_sequences | 0|none | 3|acc_norm |↑ |0.1440|± |0.0222|
|
| 111 |
+
| - leaderboard_bbh_tracking_shuffled_objects_five_objects | 0|none | 3|acc_norm |↑ |0.1560|± |0.0230|
|
| 112 |
+
| - leaderboard_bbh_tracking_shuffled_objects_seven_objects| 0|none | 3|acc_norm |↑ |0.1320|± |0.0215|
|
| 113 |
+
| - leaderboard_bbh_tracking_shuffled_objects_three_objects| 0|none | 3|acc_norm |↑ |0.2840|± |0.0286|
|
| 114 |
+
| - leaderboard_bbh_web_of_lies | 0|none | 3|acc_norm |↑ |0.4840|± |0.0317|
|
| 115 |
+
| - leaderboard_gpqa |N/A |none | 0|acc_norm |↑ |0.2903|± |0.0132|
|
| 116 |
+
| - leaderboard_gpqa_diamond | 1|none | 0|acc_norm |↑ |0.2980|± |0.0326|
|
| 117 |
+
| - leaderboard_gpqa_extended | 1|none | 0|acc_norm |↑ |0.2839|± |0.0193|
|
| 118 |
+
| - leaderboard_gpqa_main | 1|none | 0|acc_norm |↑ |0.2946|± |0.0216|
|
| 119 |
+
| - leaderboard_ifeval | 2|none | 0|inst_level_loose_acc |↑ |0.3825|± |N/A |
|
| 120 |
+
| | |none | 0|inst_level_strict_acc |↑ |0.3597|± |N/A |
|
| 121 |
+
| | |none | 0|prompt_level_loose_acc |↑ |0.2421|± |0.0184|
|
| 122 |
+
| | |none | 0|prompt_level_strict_acc|↑ |0.2181|± |0.0178|
|
| 123 |
+
| - leaderboard_math_algebra_hard | 1|none | 4|exact_match |↑ |0.1596|± |0.0209|
|
| 124 |
+
| - leaderboard_math_counting_and_prob_hard | 1|none | 4|exact_match |↑ |0.0488|± |0.0195|
|
| 125 |
+
| - leaderboard_math_geometry_hard | 1|none | 4|exact_match |↑ |0.0530|± |0.0196|
|
| 126 |
+
| - leaderboard_math_hard |N/A |none | 4|exact_match |↑ |0.0982|± |0.0079|
|
| 127 |
+
| - leaderboard_math_intermediate_algebra_hard | 1|none | 4|exact_match |↑ |0.0143|± |0.0071|
|
| 128 |
+
| - leaderboard_math_num_theory_hard | 1|none | 4|exact_match |↑ |0.0455|± |0.0168|
|
| 129 |
+
| - leaderboard_math_prealgebra_hard | 1|none | 4|exact_match |↑ |0.2591|± |0.0316|
|
| 130 |
+
| - leaderboard_math_precalculus_hard | 1|none | 4|exact_match |↑ |0.0519|± |0.0192|
|
| 131 |
+
| - leaderboard_mmlu_pro | 0.1|none | 5|acc |↑ |0.2926|± |0.0041|
|
| 132 |
+
| - leaderboard_musr |N/A |none | 0|acc_norm |↑ |0.3862|± |0.0173|
|
| 133 |
+
| - leaderboard_musr_murder_mysteries | 1|none | 0|acc_norm |↑ |0.5280|± |0.0316|
|
| 134 |
+
| - leaderboard_musr_object_placements | 1|none | 0|acc_norm |↑ |0.3594|± |0.0300|
|
| 135 |
+
| - leaderboard_musr_team_allocation | 1|none | 0|acc_norm |↑ |0.2720|± |0.0282|
|
| 136 |
+
|
| 137 |
+
| Groups |Version|Filter|n-shot| Metric | |Value | |Stderr|
|
| 138 |
+
|------------------------|-------|------|-----:|-----------------------|---|-----:|---|------|
|
| 139 |
+
|leaderboard |N/A |none | 0|acc |↑ |0.2926|± |0.0041|
|
| 140 |
+
| | |none | 0|acc_norm |↑ |0.4513|± |0.0053|
|
| 141 |
+
| | |none | 0|exact_match |↑ |0.0982|± |0.0079|
|
| 142 |
+
| | |none | 0|inst_level_loose_acc |↑ |0.3825|± |N/A |
|
| 143 |
+
| | |none | 0|inst_level_strict_acc |↑ |0.3597|± |N/A |
|
| 144 |
+
| | |none | 0|prompt_level_loose_acc |↑ |0.2421|± |0.0184|
|
| 145 |
+
| | |none | 0|prompt_level_strict_acc|↑ |0.2181|± |0.0178|
|
| 146 |
+
| - leaderboard_bbh |N/A |none | 3|acc_norm |↑ |0.4931|± |0.0061|
|
| 147 |
+
| - leaderboard_gpqa |N/A |none | 0|acc_norm |↑ |0.2903|± |0.0132|
|
| 148 |
+
| - leaderboard_math_hard|N/A |none | 4|exact_match |↑ |0.0982|± |0.0079|
|
| 149 |
+
| - leaderboard_musr |N/A |none | 0|acc_norm |↑ |0.3862|± |0.0173|
|
| 150 |
+
```
|
| 151 |
+
|
| 152 |
+
</details>
|
| 153 |
+
|
| 154 |
+
[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
|
| 155 |
+
<details><summary>See axolotl config</summary>
|
| 156 |
+
|
| 157 |
+
axolotl version: `0.4.1`
|
| 158 |
+
```yaml
|
| 159 |
+
base_model: meta-llama/Meta-Llama-3.1-8B
|
| 160 |
+
model_type: LlamaForCausalLM
|
| 161 |
+
tokenizer_type: AutoTokenizer
|
| 162 |
+
|
| 163 |
+
load_in_8bit: false
|
| 164 |
+
# load_in_4bit: true
|
| 165 |
+
strict: false
|
| 166 |
+
|
| 167 |
+
datasets:
|
| 168 |
+
- path: /workspace/datasets/dolphin-2.9.4/dolphin201-sharegpt2.jsonl
|
| 169 |
+
type: sharegpt
|
| 170 |
+
conversation: chatml
|
| 171 |
+
|
| 172 |
+
chat_template: chatml
|
| 173 |
+
# adapter: qlora
|
| 174 |
+
# lora_r: 128
|
| 175 |
+
# lora_alpha: 16
|
| 176 |
+
# lora_modules_to_save: [embed_tokens, lm_head]
|
| 177 |
+
# lora_dropout: 0.05
|
| 178 |
+
# lora_target_linear: true
|
| 179 |
+
|
| 180 |
+
unfrozen_parameters:
|
| 181 |
+
- input_layernorm
|
| 182 |
+
- model.norm
|
| 183 |
+
- post_attention_layernorm
|
| 184 |
+
- self_attn.rotary_emb
|
| 185 |
+
- ^lm_head.weight$
|
| 186 |
+
- ^model.embed_tokens.weight$
|
| 187 |
+
# mlp.down_proj layers
|
| 188 |
+
- model.layers.1.mlp.down_proj
|
| 189 |
+
- model.layers.0.mlp.down_proj
|
| 190 |
+
- model.layers.30.mlp.down_proj
|
| 191 |
+
- model.layers.2.mlp.down_proj
|
| 192 |
+
- model.layers.21.mlp.down_proj
|
| 193 |
+
- model.layers.22.mlp.down_proj
|
| 194 |
+
- model.layers.29.mlp.down_proj
|
| 195 |
+
- model.layers.5.mlp.down_proj
|
| 196 |
+
- model.layers.4.mlp.down_proj
|
| 197 |
+
- model.layers.20.mlp.down_proj
|
| 198 |
+
- model.layers.23.mlp.down_proj
|
| 199 |
+
- model.layers.19.mlp.down_proj
|
| 200 |
+
- model.layers.3.mlp.down_proj
|
| 201 |
+
- model.layers.17.mlp.down_proj
|
| 202 |
+
- model.layers.6.mlp.down_proj
|
| 203 |
+
- model.layers.31.mlp.down_proj
|
| 204 |
+
# mlp.up_proj layers
|
| 205 |
+
- model.layers.4.mlp.up_proj
|
| 206 |
+
- model.layers.3.mlp.up_proj
|
| 207 |
+
- model.layers.0.mlp.up_proj
|
| 208 |
+
- model.layers.5.mlp.up_proj
|
| 209 |
+
- model.layers.7.mlp.up_proj
|
| 210 |
+
- model.layers.6.mlp.up_proj
|
| 211 |
+
- model.layers.2.mlp.up_proj
|
| 212 |
+
- model.layers.1.mlp.up_proj
|
| 213 |
+
- model.layers.8.mlp.up_proj
|
| 214 |
+
- model.layers.12.mlp.up_proj
|
| 215 |
+
- model.layers.14.mlp.up_proj
|
| 216 |
+
- model.layers.9.mlp.up_proj
|
| 217 |
+
- model.layers.15.mlp.up_proj
|
| 218 |
+
- model.layers.17.mlp.up_proj
|
| 219 |
+
- model.layers.13.mlp.up_proj
|
| 220 |
+
- model.layers.19.mlp.up_proj
|
| 221 |
+
# self_attn.k_proj layers
|
| 222 |
+
- model.layers.29.self_attn.k_proj
|
| 223 |
+
- model.layers.25.self_attn.k_proj
|
| 224 |
+
- model.layers.23.self_attn.k_proj
|
| 225 |
+
- model.layers.28.self_attn.k_proj
|
| 226 |
+
- model.layers.21.self_attn.k_proj
|
| 227 |
+
- model.layers.19.self_attn.k_proj
|
| 228 |
+
- model.layers.22.self_attn.k_proj
|
| 229 |
+
- model.layers.20.self_attn.k_proj
|
| 230 |
+
- model.layers.24.self_attn.k_proj
|
| 231 |
+
- model.layers.31.self_attn.k_proj
|
| 232 |
+
- model.layers.27.self_attn.k_proj
|
| 233 |
+
- model.layers.26.self_attn.k_proj
|
| 234 |
+
- model.layers.17.self_attn.k_proj
|
| 235 |
+
- model.layers.11.self_attn.k_proj
|
| 236 |
+
- model.layers.18.self_attn.k_proj
|
| 237 |
+
- model.layers.14.self_attn.k_proj
|
| 238 |
+
# self_attn.o_proj layers
|
| 239 |
+
- model.layers.14.self_attn.o_proj
|
| 240 |
+
- model.layers.7.self_attn.o_proj
|
| 241 |
+
- model.layers.5.self_attn.o_proj
|
| 242 |
+
- model.layers.11.self_attn.o_proj
|
| 243 |
+
- model.layers.6.self_attn.o_proj
|
| 244 |
+
- model.layers.24.self_attn.o_proj
|
| 245 |
+
- model.layers.9.self_attn.o_proj
|
| 246 |
+
- model.layers.13.self_attn.o_proj
|
| 247 |
+
- model.layers.10.self_attn.o_proj
|
| 248 |
+
- model.layers.12.self_attn.o_proj
|
| 249 |
+
- model.layers.8.self_attn.o_proj
|
| 250 |
+
- model.layers.25.self_attn.o_proj
|
| 251 |
+
- model.layers.21.self_attn.o_proj
|
| 252 |
+
- model.layers.23.self_attn.o_proj
|
| 253 |
+
- model.layers.15.self_attn.o_proj
|
| 254 |
+
- model.layers.16.self_attn.o_proj
|
| 255 |
+
# self_attn.q_proj layers
|
| 256 |
+
- model.layers.8.self_attn.q_proj
|
| 257 |
+
- model.layers.13.self_attn.q_proj
|
| 258 |
+
- model.layers.9.self_attn.q_proj
|
| 259 |
+
- model.layers.14.self_attn.q_proj
|
| 260 |
+
- model.layers.10.self_attn.q_proj
|
| 261 |
+
- model.layers.11.self_attn.q_proj
|
| 262 |
+
- model.layers.0.self_attn.q_proj
|
| 263 |
+
- model.layers.15.self_attn.q_proj
|
| 264 |
+
- model.layers.1.self_attn.q_proj
|
| 265 |
+
- model.layers.6.self_attn.q_proj
|
| 266 |
+
- model.layers.5.self_attn.q_proj
|
| 267 |
+
- model.layers.7.self_attn.q_proj
|
| 268 |
+
- model.layers.12.self_attn.q_proj
|
| 269 |
+
- model.layers.16.self_attn.q_proj
|
| 270 |
+
- model.layers.17.self_attn.q_proj
|
| 271 |
+
- model.layers.26.self_attn.q_proj
|
| 272 |
+
# self_attn.v_proj layers
|
| 273 |
+
- model.layers.26.self_attn.v_proj
|
| 274 |
+
- model.layers.17.self_attn.v_proj
|
| 275 |
+
- model.layers.3.self_attn.v_proj
|
| 276 |
+
- model.layers.28.self_attn.v_proj
|
| 277 |
+
- model.layers.29.self_attn.v_proj
|
| 278 |
+
- model.layers.21.self_attn.v_proj
|
| 279 |
+
- model.layers.15.self_attn.v_proj
|
| 280 |
+
- model.layers.16.self_attn.v_proj
|
| 281 |
+
- model.layers.20.self_attn.v_proj
|
| 282 |
+
- model.layers.25.self_attn.v_proj
|
| 283 |
+
- model.layers.6.self_attn.v_proj
|
| 284 |
+
- model.layers.23.self_attn.v_proj
|
| 285 |
+
- model.layers.4.self_attn.v_proj
|
| 286 |
+
- model.layers.1.self_attn.v_proj
|
| 287 |
+
- model.layers.22.self_attn.v_proj
|
| 288 |
+
- model.layers.14.self_attn.v_proj
|
| 289 |
+
# mlp.gate_proj layers
|
| 290 |
+
- model.layers.1.mlp.gate_proj
|
| 291 |
+
- model.layers.2.mlp.gate_proj
|
| 292 |
+
- model.layers.3.mlp.gate_proj
|
| 293 |
+
- model.layers.4.mlp.gate_proj
|
| 294 |
+
- model.layers.0.mlp.gate_proj
|
| 295 |
+
- model.layers.25.mlp.gate_proj
|
| 296 |
+
- model.layers.26.mlp.gate_proj
|
| 297 |
+
- model.layers.5.mlp.gate_proj
|
| 298 |
+
- model.layers.24.mlp.gate_proj
|
| 299 |
+
- model.layers.28.mlp.gate_proj
|
| 300 |
+
- model.layers.23.mlp.gate_proj
|
| 301 |
+
- model.layers.27.mlp.gate_proj
|
| 302 |
+
- model.layers.21.mlp.gate_proj
|
| 303 |
+
- model.layers.22.mlp.gate_proj
|
| 304 |
+
- model.layers.29.mlp.gate_proj
|
| 305 |
+
- model.layers.20.mlp.gate_proj
|
| 306 |
+
|
| 307 |
+
|
| 308 |
+
|
| 309 |
+
|
| 310 |
+
dataset_prepared_path: /workspace/axolotl/dolph-2.9.4-nemo-prepared
|
| 311 |
+
val_set_size: 0.01
|
| 312 |
+
output_dir: /workspace/axolotl/dolphin-2.9.4-llama3.1-8b
|
| 313 |
+
|
| 314 |
+
sequence_len: 8192
|
| 315 |
+
sample_packing: true
|
| 316 |
+
pad_to_sequence_len: true
|
| 317 |
+
|
| 318 |
+
wandb_project: dolphin-2.9.4-llama3.1-8b
|
| 319 |
+
wandb_watch:
|
| 320 |
+
wandb_run_id:
|
| 321 |
+
wandb_log_model:
|
| 322 |
+
|
| 323 |
+
gradient_accumulation_steps: 16
|
| 324 |
+
micro_batch_size: 2
|
| 325 |
+
num_epochs: 3
|
| 326 |
+
optimizer: adamw_torch
|
| 327 |
+
lr_scheduler: cosine
|
| 328 |
+
learning_rate: 5e-6
|
| 329 |
+
train_on_inputs: false
|
| 330 |
+
group_by_length: false
|
| 331 |
+
bf16: auto
|
| 332 |
+
fp16:
|
| 333 |
+
tf32:
|
| 334 |
+
|
| 335 |
+
gradient_checkpointing: true
|
| 336 |
+
gradient_checkpointing_kwargs:
|
| 337 |
+
use_reentrant: false
|
| 338 |
+
early_stopping_patience:
|
| 339 |
+
resume_from_checkpoint:
|
| 340 |
+
logging_steps: 1
|
| 341 |
+
xformers_attention:
|
| 342 |
+
flash_attention: true
|
| 343 |
+
|
| 344 |
+
warmup_steps: 100
|
| 345 |
+
# evals_per_epoch: 4
|
| 346 |
+
eval_table_size:
|
| 347 |
+
saves_per_epoch: 1
|
| 348 |
+
save_total_limit: 2
|
| 349 |
+
save_steps:
|
| 350 |
+
debug:
|
| 351 |
+
deepspeed: deepspeed_configs/zero3_bf16.json
|
| 352 |
+
weight_decay: 0.1
|
| 353 |
+
special_tokens:
|
| 354 |
+
eos_token: "<|im_end|>"
|
| 355 |
+
bos_token: "<|begin_of_text|>"
|
| 356 |
+
pad_token: "<|finetune_right_pad_id|>"
|
| 357 |
+
tokens:
|
| 358 |
+
- "<|im_start|>"
|
| 359 |
+
|
| 360 |
+
|
| 361 |
+
# fsdp:
|
| 362 |
+
# - full_shard
|
| 363 |
+
# - auto_wrap
|
| 364 |
+
# fsdp_config:
|
| 365 |
+
# fsdp_limit_all_gathers: true
|
| 366 |
+
# fsdp_sync_module_states: true
|
| 367 |
+
# fsdp_offload_params: true
|
| 368 |
+
# fsdp_use_orig_params: false
|
| 369 |
+
# fsdp_cpu_ram_efficient_loading: true
|
| 370 |
+
# fsdp_transformer_layer_cls_to_wrap: MixtralSparseMoeBlock
|
| 371 |
+
# fsdp_state_dict_type: FULL_STATE_DICT
|
| 372 |
+
# fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
|
| 373 |
+
# fsdp_sharding_strategy: FULL_SHARD
|
| 374 |
+
# fsdp_forward_prefetch: false
|
| 375 |
+
# fsdp_backward_prefetch: BACKWARD_PRE
|
| 376 |
+
```
|
| 377 |
+
|
| 378 |
+
</details><br>
|
| 379 |
+
|
| 380 |
+
# workspace/axolotl/dolphin-2.9.4-llama3.1-8b
|
| 381 |
+
|
| 382 |
+
This model is a fine-tuned version of [meta-llama/Meta-Llama-3.1-8B](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B) on the None dataset.
|
| 383 |
+
It achieves the following results on the evaluation set:
|
| 384 |
+
- Loss: 0.5655
|
| 385 |
+
|
| 386 |
+
## Model description
|
| 387 |
+
|
| 388 |
+
More information needed
|
| 389 |
+
|
| 390 |
+
## Intended uses & limitations
|
| 391 |
+
|
| 392 |
+
More information needed
|
| 393 |
+
|
| 394 |
+
## Training and evaluation data
|
| 395 |
+
|
| 396 |
+
More information needed
|
| 397 |
+
|
| 398 |
+
## Training procedure
|
| 399 |
+
|
| 400 |
+
### Training hyperparameters
|
| 401 |
+
|
| 402 |
+
The following hyperparameters were used during training:
|
| 403 |
+
- learning_rate: 5e-06
|
| 404 |
+
- train_batch_size: 2
|
| 405 |
+
- eval_batch_size: 2
|
| 406 |
+
- seed: 42
|
| 407 |
+
- distributed_type: multi-GPU
|
| 408 |
+
- num_devices: 8
|
| 409 |
+
- gradient_accumulation_steps: 16
|
| 410 |
+
- total_train_batch_size: 256
|
| 411 |
+
- total_eval_batch_size: 16
|
| 412 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 413 |
+
- lr_scheduler_type: cosine
|
| 414 |
+
- lr_scheduler_warmup_steps: 100
|
| 415 |
+
- num_epochs: 3
|
| 416 |
+
|
| 417 |
+
### Training results
|
| 418 |
+
|
| 419 |
+
| Training Loss | Epoch | Step | Validation Loss |
|
| 420 |
+
|:-------------:|:------:|:----:|:---------------:|
|
| 421 |
+
| 0.5837 | 1.0180 | 1161 | 0.5814 |
|
| 422 |
+
| 0.5525 | 2.0179 | 2322 | 0.5671 |
|
| 423 |
+
| 0.5514 | 2.9624 | 3420 | 0.5655 |
|
| 424 |
+
|
| 425 |
+
|
| 426 |
+
### Framework versions
|
| 427 |
+
|
| 428 |
+
- Transformers 4.44.0.dev0
|
| 429 |
+
- Pytorch 2.4.0+cu121
|
| 430 |
+
- Datasets 2.19.1
|
| 431 |
+
- Tokenizers 0.19.1
|