Delta-Vector
commited on
Commit
•
c6ba0c1
1
Parent(s):
5393fff
Delete README.md
Browse files
README.md
DELETED
@@ -1,166 +0,0 @@
|
|
1 |
-
---
|
2 |
-
License: apache-2.0
|
3 |
-
Language:
|
4 |
-
- En
|
5 |
-
Pipeline_tag: text-generation
|
6 |
-
Base_model: nvidia/Llama-3.1-Minitron-4 B-Width-Base
|
7 |
-
Tags:
|
8 |
-
- Chat
|
9 |
-
license: agpl-3.0
|
10 |
-
datasets:
|
11 |
-
- anthracite-org/kalo-opus-instruct-22k-no-refusal
|
12 |
-
- PJMixers/lodrick-the-lafted_OpusStories-ShareGPT
|
13 |
-
- NewEden/Gryphe-3.5-16k-Subset
|
14 |
-
- Epiculous/Synthstruct-Gens-v1.1-Filtered-n-Cleaned
|
15 |
-
tags:
|
16 |
-
- chat
|
17 |
-
---
|
18 |
-
|
19 |
-
![image/png](https://huggingface.co/Edens-Gate/Testing123/resolve/main/oie_gM9EsNXjMDsT.jpg?download=true)
|
20 |
-
A model made to continue off my previous work on [Magnum 4B](https://huggingface.co/anthracite-org/magnum-v2-4b), A small model made for creative writing / General assistant tasks, finetuned ontop of [IntervitensInc/Llama-3.1-Minitron-4B-Width-Base-chatml](https://huggingface.co/IntervitensInc/Llama-3.1-Minitron-4B-Width-Base-chatml), this model is made to be more coherent and generally be better then the 4B at both writing and assistant tasks.
|
21 |
-
|
22 |
-
## Prompting
|
23 |
-
Model has been Instruct tuned with the ChatML formatting. A typical input would look like this:
|
24 |
-
|
25 |
-
```py
|
26 |
-
"""<|im_start|>system
|
27 |
-
system prompt<|im_end|>
|
28 |
-
<|im_start|>user
|
29 |
-
Hi there!<|im_end|>
|
30 |
-
<|im_start|>assistant
|
31 |
-
Nice to meet you!<|im_end|>
|
32 |
-
<|im_start|>user
|
33 |
-
Can I ask a question?<|im_end|>
|
34 |
-
<|im_start|>assistant
|
35 |
-
"""
|
36 |
-
```
|
37 |
-
|
38 |
-
## Support
|
39 |
-
|
40 |
-
To run inference on this model, you'll need to use Aphrodite, vLLM or EXL 2/tabbyAPI, as llama.cpp hasn't yet merged the required pull request to fix the llama 3.1 rope_freqs issue with custom head dimensions.
|
41 |
-
|
42 |
-
However, you can work around this by quantizing the model yourself to create a functional GGUF file. Note that until [this PR](https://github.com/ggerganov/llama.cpp/pull/9141) is merged, the context will be limited to 8 k tokens.
|
43 |
-
|
44 |
-
To create a working GGUF file, make the following adjustments:
|
45 |
-
|
46 |
-
1. Remove the `"rope_scaling": {}` entry from `config.json`
|
47 |
-
2. Change `"max_position_embeddings"` to `8192` in `config.json`
|
48 |
-
|
49 |
-
These modifications should allow you to use the model with llama. Cpp, albeit with the mentioned context limitation.
|
50 |
-
|
51 |
-
## Axolotl config
|
52 |
-
|
53 |
-
<details><summary>See axolotl config</summary>
|
54 |
-
|
55 |
-
Axolotl version: `0.4.1`
|
56 |
-
```yaml
|
57 |
-
base_model: IntervitensInc/Llama-3.1-Minitron-4B-Width-Base-chatml
|
58 |
-
model_type: AutoModelForCausalLM
|
59 |
-
tokenizer_type: AutoTokenizer
|
60 |
-
|
61 |
-
load_in_8bit: false
|
62 |
-
load_in_4bit: false
|
63 |
-
strict: false
|
64 |
-
|
65 |
-
datasets:
|
66 |
-
- path: NewEden/Gryphe-3.5-16k-Subset
|
67 |
-
type: sharegpt
|
68 |
-
conversation: chatml
|
69 |
-
- path: Epiculous/Synthstruct-Gens-v1.1-Filtered-n-Cleaned
|
70 |
-
type: sharegpt
|
71 |
-
conversation: chatml
|
72 |
-
- path: anthracite-org/kalo-opus-instruct-22k-no-refusal
|
73 |
-
type: sharegpt
|
74 |
-
conversation: chatml
|
75 |
-
- path: PJMixers/lodrick-the-lafted_OpusStories-ShareGPT
|
76 |
-
type: sharegpt
|
77 |
-
conversation: chatml
|
78 |
-
|
79 |
-
chat_template: chatml
|
80 |
-
|
81 |
-
val_set_size: 0.01
|
82 |
-
output_dir: ./outputs/out
|
83 |
-
|
84 |
-
adapter:
|
85 |
-
lora_r:
|
86 |
-
lora_alpha:
|
87 |
-
lora_dropout:
|
88 |
-
lora_target_linear:
|
89 |
-
|
90 |
-
sequence_len: 16384
|
91 |
-
# sequence_len: 32768
|
92 |
-
sample_packing: true
|
93 |
-
eval_sample_packing: false
|
94 |
-
pad_to_sequence_len: true
|
95 |
-
|
96 |
-
plugins:
|
97 |
-
- axolotl.integrations.liger.LigerPlugin
|
98 |
-
liger_rope: true
|
99 |
-
liger_rms_norm: true
|
100 |
-
liger_swiglu: true
|
101 |
-
liger_fused_linear_cross_entropy: true
|
102 |
-
|
103 |
-
wandb_project:
|
104 |
-
wandb_entity:
|
105 |
-
wandb_watch:
|
106 |
-
wandb_name:
|
107 |
-
wandb_log_model:
|
108 |
-
|
109 |
-
gradient_accumulation_steps: 32
|
110 |
-
micro_batch_size: 1
|
111 |
-
num_epochs: 2
|
112 |
-
optimizer: adamw_bnb_8bit
|
113 |
-
#optimizer: paged_adamw_8bit
|
114 |
-
lr_scheduler: cosine
|
115 |
-
learning_rate: 0.00002
|
116 |
-
weight_decay: 0.05
|
117 |
-
|
118 |
-
train_on_inputs: false
|
119 |
-
group_by_length: false
|
120 |
-
bf16: auto
|
121 |
-
fp16:
|
122 |
-
tf32: true
|
123 |
-
|
124 |
-
gradient_checkpointing: true
|
125 |
-
early_stopping_patience:
|
126 |
-
resume_from_checkpoint:
|
127 |
-
local_rank:
|
128 |
-
logging_steps: 1
|
129 |
-
xformers_attention:
|
130 |
-
flash_attention: true
|
131 |
-
|
132 |
-
warmup_ratio: 0.1
|
133 |
-
evals_per_epoch: 4
|
134 |
-
eval_table_size:
|
135 |
-
eval_max_new_tokens: 128
|
136 |
-
saves_per_epoch: 1
|
137 |
-
|
138 |
-
debug:
|
139 |
-
deepspeed: /workspace/axolotl/deepspeed_configs/zero2.json
|
140 |
-
#deepspeed:
|
141 |
-
fsdp:
|
142 |
-
fsdp_config:
|
143 |
-
|
144 |
-
special_tokens:
|
145 |
-
pad_token: <|finetune_right_pad_id|>
|
146 |
-
|
147 |
-
```
|
148 |
-
|
149 |
-
</details><br>
|
150 |
-
|
151 |
-
## Credits
|
152 |
-
|
153 |
-
- [anthracite-org/kalo-opus-instruct-22k-no-refusal](https://huggingface.co/datasets/anthracite-org/kalo-opus-instruct-22k-no-refusal)
|
154 |
-
- [NewEden/Gryphe-3.5-16k-Subset](https://huggingface.co/datasets/NewEden/Gryphe-3.5-16k-Subset)
|
155 |
-
- [Epiculous/Synthstruct-Gens-v1.1-Filtered-n-Cleaned](https://huggingface.co/datasets/Epiculous/Synthstruct-Gens-v1.1-Filtered-n-Cleaned)
|
156 |
-
- [lodrick-the-lafted/OpusStories](https://huggingface.co/datasets/lodrick-the-lafted/OpusStories)
|
157 |
-
|
158 |
-
I couldn't have made this model without the help of [Kubernetes_bad](https://huggingface.co/kubernetes-bad) and the support of [Lucy Knada](https://huggingface.co/lucyknada)
|
159 |
-
|
160 |
-
## Training
|
161 |
-
The training was done for 2 epochs. We used 2 x [RTX 6000s](https://store.nvidia.com/en-us/nvidia-rtx/products/nvidia-rtx-6000-ada-generation/) GPUs graciously provided by [Kubernetes_Bad](https://huggingface.co/kubernetes-bad) for the full-parameter fine-tuning of the model.
|
162 |
-
|
163 |
-
[<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)
|
164 |
-
|
165 |
-
## Safety
|
166 |
-
...
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|