Views
No views yet
0.11.0.dev01base_model: meta-llama/Llama-3.1-70B
2base_model_ignore_patterns: "*/*"
3model_type: AutoModelForCausalLM
4tokenizer_type: AutoTokenizer
5
6load_in_8bit: false
7load_in_4bit: false
8strict: false
9
10datasets:
11 - path: merged.json
12 type: dan-chat-advanced-llama3
13 - path: PocketDoc/Dans-Prosemaxx-RP
14 type: dan-chat-advanced-llama3
15 - path: PocketDoc/Dans-Personamaxx-Logs-2
16 type: dan-chat-advanced-llama3
17 - path: PocketDoc/Dans-Personamaxx-VN
18 type: dan-chat-advanced-llama3
19 - path: PocketDoc/Dans-Kinomaxx-VanillaBackrooms
20 type: dan-chat-advanced-llama3
21 - path: PocketDoc/Dans-Prosemaxx-Gutenberg
22 type: dan-chat-advanced-llama3
23 - path: PocketDoc/Dans-Prosemaxx-Cowriter-3-XL
24 type: dan-chat-advanced-llama3
25 - path: PocketDoc/Dans-Prosemaxx-Adventure
26 type: dan-chat-advanced-llama3
27 - path: PocketDoc/Dans-Failuremaxx-Adventure-3
28 type: dan-chat-advanced-llama3
29 - path: PocketDoc/Dans-Prosemaxx-InstructWriter-ZeroShot-2
30 type: dan-chat-advanced-llama3
31 - path: PocketDoc/Dans-Prosemaxx-InstructWriter-ZeroShot-3
32 type: dan-chat-advanced-llama3
33 - path: PocketDoc/Dans-Prosemaxx-InstructWriter-Continue-2
34 type: dan-chat-advanced-llama3
35 - path: PocketDoc/Dans-Prosemaxx-Instructwriter-Long
36 type: dan-chat-advanced-llama3
37 - path: PocketDoc/Dans-Prosemaxx-RepRemover-1
38 type: dan-chat-advanced-llama3
39 - path: PocketDoc/Dans-MemoryCore-CoreCurriculum-Small
40 type: dan-chat-advanced-llama3
41 - path: AquaV/US-Army-Survival-Sharegpt
42 type: dan-chat-advanced-llama3
43 - path: AquaV/Multi-Environment-Operations-Sharegpt
44 type: dan-chat-advanced-llama3
45 - path: AquaV/Resistance-Sharegpt
46 type: dan-chat-advanced-llama3
47 - path: AquaV/Interrogation-Sharegpt
48 type: dan-chat-advanced-llama3
49 - path: AquaV/Chemical-Biological-Safety-Applications-Sharegpt
50 type: dan-chat-advanced-llama3
51 - path: AquaV/Energetic-Materials-Sharegpt
52 type: dan-chat-advanced-llama3
53 - path: PocketDoc/Dans-Mathmaxx
54 type: dan-chat-advanced-llama3
55 - path: PJMixers/Math-Multiturn-1K-ShareGPT
56 type: dan-chat-advanced-llama3
57 - path: PocketDoc/Dans-Taskmaxx
58 type: dan-chat-advanced-llama3
59 - path: PocketDoc/Dans-Taskmaxx-DataPrepper
60 type: dan-chat-advanced-llama3
61 - path: PocketDoc/Dans-Taskmaxx-ConcurrentQA-Reworked
62 type: dan-chat-advanced-llama3
63 - path: PocketDoc/Dans-Taskmaxx-TableGPT
64 type: dan-chat-advanced-llama3
65 - path: PocketDoc/Dans-Taskmaxx-SciRIFF
66 type: dan-chat-advanced-llama3
67 - path: PocketDoc/Dans-Taskmaxx-Edit
68 type: dan-chat-advanced-llama3
69 - path: PocketDoc/Dans-Toolmaxx-Agent
70 type: dan-chat-advanced-llama3
71 - path: PocketDoc/Dans-Toolmaxx-ShellCommands
72 type: dan-chat-advanced-llama3
73 - path: PocketDoc/Dans-Toolmaxx-Functions-Toolbench
74 type: dan-chat-advanced-llama3
75 - path: PocketDoc/Dans-Toolmaxx-Functions-ToolACE
76 type: dan-chat-advanced-llama3
77 - path: PocketDoc/Dans-Toolmaxx-Functions-apigen-subset
78 type: dan-chat-advanced-llama3
79 - path: PocketDoc/Dans-Assistantmaxx-OpenAssistant2
80 type: dan-chat-advanced-llama3
81 - path: PocketDoc/Dans-Assistantmaxx-Opus-Merge-2
82 type: dan-chat-advanced-llama3
83 - path: PocketDoc/Dans-Assistantmaxx-sonnetorca-subset
84 type: dan-chat-advanced-llama3
85 - path: PocketDoc/Dans-Assistantmaxx-sonnetorca-subset-2
86 type: dan-chat-advanced-llama3
87 - path: PocketDoc/Dans-Assistantmaxx-Synthia
88 type: dan-chat-advanced-llama3
89 - path: PocketDoc/Dans-Assistantmaxx-ASL
90 type: dan-chat-advanced-llama3
91 - path: PocketDoc/Dans-Assistantmaxx-PersonaLLM-Opus
92 type: dan-chat-advanced-llama3
93 - path: PocketDoc/Dans-Assistantmaxx-LongAlign
94 type: dan-chat-advanced-llama3
95 - path: PocketDoc/Dans-Assistantmaxx-OpenLeecher-Instruct
96 type: dan-chat-advanced-llama3
97 - path: PocketDoc/Dans-Assistantmaxx-Tulu3-IF
98 type: dan-chat-advanced-llama3
99 - path: PocketDoc/Dans-Systemmaxx
100 type: dan-chat-advanced-llama3
101 - path: PocketDoc/Dans-Logicmaxx-SAT-AP
102 type: dan-chat-advanced-llama3
103 - path: PJMixers/grimulkan_theory-of-mind-ShareGPT
104 type: dan-chat-advanced-llama3
105 - path: PJMixers/grimulkan_physical-reasoning-ShareGPT
106 type: dan-chat-advanced-llama3
107 - path: PocketDoc/Dans-Reasoningmaxx-NaturalReasoning
108 type: dan-chat-advanced-llama3
109 - path: PocketDoc/Dans-Reasoningmaxx-WebInstruct
110 type: dan-chat-advanced-llama3
111 - path: PocketDoc/Dans-Reasoningmaxx-GeneralReasoning
112 type: dan-chat-advanced-llama3
113 - path: Delta-Vector/Orion-LN-V1-ShareGPT
114 type: dan-chat-advanced-llama3
115 - path: Delta-Vector/Orion-Alpindale-LN-ShareGPT
116 type: dan-chat-advanced-llama3
117 - path: Delta-Vector/Orion-Shoujo-AI-Filtered-ShareGPT
118 type: dan-chat-advanced-llama3
119 - path: Delta-Vector/Orion-RP-Guild
120 type: dan-chat-advanced-llama3
121 - path: Delta-Vector/Orion-OpenCAI-ShareGPT
122 type: dan-chat-advanced-llama3
123 - path: Delta-Vector/Orion-LIMARP-Complexity
124 type: dan-chat-advanced-llama3
125shuffle_merged_datasets: true
126dataset_prepared_path: /home/ubuntu/dpe-70b-data
127val_set_size: 0.0
128output_dir: /home/ubuntu/dpe-70b-fft-rev1
129
130plugins:
131 - axolotl.integrations.liger.LigerPlugin
132 - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
133liger_rope: true
134liger_rms_norm: true
135liger_layer_norm: true
136liger_glu_activation: true
137liger_fused_linear_cross_entropy: false
138cut_cross_entropy: true
139
140sequence_len: 32768
141sample_packing: true
142eval_sample_packing: true
143pad_to_sequence_len: true
144
145wandb_project: dpe-70b-fft
146wandb_entity:
147wandb_watch:
148wandb_name:
149wandb_log_model:
150
151gradient_accumulation_steps: 4
152micro_batch_size: 1
153num_epochs: 2
154optimizer: paged_ademamix_8bit
155optim_args: "beta1=0.9,beta2=0.999,beta3=0.999,alpha=5"
156lr_scheduler: rex
157learning_rate: 1.0e-6
158max_grad_norm: 0.001
159
160train_on_inputs: false
161group_by_length: false
162bf16: auto
163fp16:
164tf32: false
165
166gradient_checkpointing: true
167early_stopping_patience:
168resume_from_checkpoint:
169local_rank:
170logging_steps: 1
171xformers_attention:
172flash_attention: true
173s2_attention:
174
175warmup_ratio: 0.1
176evals_per_epoch:
177eval_table_size:
178eval_max_new_tokens:
179saves_per_epoch: 2
180debug:
181deepspeed: ./deepspeed_configs/zero3_bf16.json
182weight_decay: 0.0025
183fsdp:
184fsdp_config:
185special_tokens:
186 eos_token: <|eot_id|>
187 pad_token: <|finetune_right_pad_id|>
188