add things I already know pre model card 2
Browse files
README.md
CHANGED
@@ -32,7 +32,81 @@ This model's training was sponsored by [sablo.ai](https://sablo.ai).
|
|
32 |
|
33 |
axolotl version: `0.4.0`
|
34 |
```yaml
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
35 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
36 |
```
|
37 |
|
38 |
</details><br>
|
@@ -75,7 +149,7 @@ Quantizationed versions of this model is currently not available. It will be ava
|
|
75 |
|
76 |
This model is full fine-tuned for 2 epoch.
|
77 |
|
78 |
-
Total number of steps was
|
79 |
|
80 |
<details><summary>Loss graph</summary>
|
81 |
|
|
|
32 |
|
33 |
axolotl version: `0.4.0`
|
34 |
```yaml
|
35 |
+
base_model: meta-math/MetaMath-Mistral-7B
|
36 |
+
model_type: MistralForCausalLM
|
37 |
+
tokenizer_type: LlamaTokenizer
|
38 |
+
is_mistral_derived_model: true
|
39 |
+
|
40 |
+
load_in_8bit: false
|
41 |
+
load_in_4bit: false
|
42 |
+
strict: false
|
43 |
|
44 |
+
chat_template: alpaca
|
45 |
+
datasets:
|
46 |
+
- path: microsoft/orca-math-word-problems-200k
|
47 |
+
type: alpaca_chat.load_qa
|
48 |
+
conversation: alpaca
|
49 |
+
|
50 |
+
- path: TIGER-Lab/MathInstruct
|
51 |
+
type: alpaca
|
52 |
+
conversation: alpaca
|
53 |
+
|
54 |
+
dataset_prepared_path: last_run_prepared
|
55 |
+
val_set_size: 0.005
|
56 |
+
#val_set_size: 0.0
|
57 |
+
|
58 |
+
output_dir: ./EulerMath-Mistral-7B-model
|
59 |
+
|
60 |
+
sequence_len: 8192
|
61 |
+
sample_packing: true
|
62 |
+
pad_to_sequence_len: true
|
63 |
+
eval_sample_packing: false
|
64 |
+
|
65 |
+
wandb_project: Euler
|
66 |
+
wandb_entity:
|
67 |
+
wandb_watch:
|
68 |
+
wandb_name:
|
69 |
+
wandb_log_model:
|
70 |
+
hub_model_id: Weyaxi/EulerMath-Mistral-7B
|
71 |
+
|
72 |
+
save_safetensors: true
|
73 |
+
|
74 |
+
gradient_accumulation_steps: 4
|
75 |
+
micro_batch_size: 2 # changed
|
76 |
+
num_epochs: 2
|
77 |
+
optimizer: adamw_bnb_8bit
|
78 |
+
lr_scheduler: cosine
|
79 |
+
learning_rate: 0.000005
|
80 |
+
|
81 |
+
train_on_inputs: false
|
82 |
+
group_by_length: false
|
83 |
+
bf16: true
|
84 |
+
fp16: false
|
85 |
+
tf32: false
|
86 |
+
|
87 |
+
gradient_checkpointing: true
|
88 |
+
early_stopping_patience:
|
89 |
+
resume_from_checkpoint:
|
90 |
+
local_rank:
|
91 |
+
logging_steps: 1
|
92 |
+
xformers_attention:
|
93 |
+
flash_attention: true
|
94 |
+
|
95 |
+
warmup_steps: 10
|
96 |
+
evals_per_epoch: 4 # changed
|
97 |
+
eval_table_size:
|
98 |
+
eval_table_max_new_tokens: 128
|
99 |
+
saves_per_epoch: 1 # changed
|
100 |
+
debug:
|
101 |
+
|
102 |
+
deepspeed: zero3_bf16.json
|
103 |
+
weight_decay: 0.0
|
104 |
+
fsdp:
|
105 |
+
fsdp_config:
|
106 |
+
special_tokens:
|
107 |
+
bos_token: "<s>"
|
108 |
+
eos_token: "</s>"
|
109 |
+
unk_token: "<unk>"
|
110 |
```
|
111 |
|
112 |
</details><br>
|
|
|
149 |
|
150 |
This model is full fine-tuned for 2 epoch.
|
151 |
|
152 |
+
Total number of steps was 544.
|
153 |
|
154 |
<details><summary>Loss graph</summary>
|
155 |
|