gronkomatic
commited on
Update README.md
Browse files
README.md
CHANGED
@@ -9,35 +9,67 @@ pipeline_tag: text-generation
|
|
9 |
---
|
10 |
# Model Card for neoncortex/mini-mistral-openhermes-2.5-chatml-test
|
11 |
|
12 |
-
A tiny Mistral model trained on teknium/OpenHermes-2.5.
|
|
|
13 |
|
14 |
## Model Details
|
15 |
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
- **Developed by:** [More Information Needed]
|
23 |
-
- **Funded by [optional]:** [More Information Needed]
|
24 |
-
- **Shared by [optional]:** [More Information Needed]
|
25 |
-
- **Model type:** [More Information Needed]
|
26 |
-
- **Language(s) (NLP):** [More Information Needed]
|
27 |
-
- **License:** [More Information Needed]
|
28 |
-
- **Finetuned from model [optional]:** [More Information Needed]
|
29 |
-
|
30 |
-
### Model Sources [optional]
|
31 |
|
32 |
-
|
33 |
|
34 |
-
|
35 |
-
|
36 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
37 |
|
38 |
## Uses
|
39 |
|
40 |
-
|
41 |
|
42 |
### Direct Use
|
43 |
|
@@ -53,7 +85,7 @@ A tiny Mistral model trained on teknium/OpenHermes-2.5. This is epoch 3/9, so it
|
|
53 |
|
54 |
### Out-of-Scope Use
|
55 |
|
56 |
-
|
57 |
|
58 |
[More Information Needed]
|
59 |
|
@@ -87,16 +119,20 @@ Use the code below to get started with the model.
|
|
87 |
|
88 |
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
|
89 |
|
90 |
-
#### Preprocessing
|
91 |
-
|
92 |
-
[More Information Needed]
|
93 |
|
|
|
94 |
|
95 |
#### Training Hyperparameters
|
96 |
|
97 |
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
|
98 |
|
99 |
-
#### Speeds, Sizes, Times
|
|
|
|
|
|
|
|
|
|
|
100 |
|
101 |
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
|
102 |
|
@@ -104,41 +140,11 @@ Use the code below to get started with the model.
|
|
104 |
|
105 |
## Evaluation
|
106 |
|
107 |
-
|
108 |
-
|
109 |
-
### Testing Data, Factors & Metrics
|
110 |
|
111 |
-
|
112 |
|
113 |
-
|
114 |
-
|
115 |
-
[More Information Needed]
|
116 |
-
|
117 |
-
#### Factors
|
118 |
-
|
119 |
-
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
|
120 |
-
|
121 |
-
[More Information Needed]
|
122 |
-
|
123 |
-
#### Metrics
|
124 |
-
|
125 |
-
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
|
126 |
-
|
127 |
-
[More Information Needed]
|
128 |
-
|
129 |
-
### Results
|
130 |
-
|
131 |
-
[More Information Needed]
|
132 |
-
|
133 |
-
#### Summary
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
## Model Examination [optional]
|
138 |
-
|
139 |
-
<!-- Relevant interpretability work for the model goes here -->
|
140 |
-
|
141 |
-
[More Information Needed]
|
142 |
|
143 |
## Environmental Impact
|
144 |
|
@@ -146,13 +152,13 @@ Use the code below to get started with the model.
|
|
146 |
|
147 |
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
|
148 |
|
149 |
-
- **Hardware Type:**
|
150 |
-
- **Hours used:**
|
151 |
-
- **Cloud Provider:**
|
152 |
-
- **Compute Region:**
|
153 |
-
- **Carbon Emitted:**
|
154 |
|
155 |
-
## Technical Specifications
|
156 |
|
157 |
### Model Architecture and Objective
|
158 |
|
@@ -160,42 +166,28 @@ Carbon emissions can be estimated using the [Machine Learning Impact calculator]
|
|
160 |
|
161 |
### Compute Infrastructure
|
162 |
|
163 |
-
|
164 |
|
165 |
#### Hardware
|
166 |
|
167 |
-
|
168 |
|
169 |
#### Software
|
170 |
|
171 |
-
[
|
172 |
-
|
173 |
-
## Citation [optional]
|
174 |
-
|
175 |
-
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
|
176 |
|
177 |
-
|
178 |
|
179 |
-
|
180 |
|
181 |
-
|
182 |
|
183 |
[More Information Needed]
|
184 |
|
185 |
-
##
|
186 |
-
|
187 |
-
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
|
188 |
-
|
189 |
-
[More Information Needed]
|
190 |
|
191 |
-
|
192 |
-
|
193 |
-
[More Information Needed]
|
194 |
-
|
195 |
-
## Model Card Authors [optional]
|
196 |
-
|
197 |
-
[More Information Needed]
|
198 |
|
199 |
## Model Card Contact
|
200 |
|
201 |
-
|
|
|
9 |
---
|
10 |
# Model Card for neoncortex/mini-mistral-openhermes-2.5-chatml-test
|
11 |
|
12 |
+
A tiny Mistral model trained on teknium/OpenHermes-2.5.
|
13 |
+
This is epoch 5/9, so still some training to go.
|
14 |
|
15 |
## Model Details
|
16 |
|
17 |
+
A 63M parameter auto-regressive LM using Mistral architecture as a base.
|
18 |
+
- Multi-query Attention instead of Grouped-query Attention.
|
19 |
+
- Sliding window is disabled.
|
20 |
+
- Modified ChatML instead of Mistral chat template - TL;DR I used '<|im_start|>human' instead of '<|im_start|>user'
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
|
22 |
+
### Model Description
|
23 |
|
24 |
+
Just doing it to see what happens.
|
25 |
+
|
26 |
+
It'll take about 40 to 45 hours to train on two Nvidia RTX 3060 12GB.
|
27 |
+
|
28 |
+
It uses ChatML for the chat template, but I fucked up the template in the dataset,
|
29 |
+
using '<|im_start|>human' instead of '<|im_start|>user'. ¯\_(ツ)_/¯
|
30 |
+
So, here's the bits:
|
31 |
+
|
32 |
+
```
|
33 |
+
{%- set ns = namespace(found=false) -%}
|
34 |
+
{%- for message in messages -%}
|
35 |
+
{%- if message['role'] == 'system' -%}
|
36 |
+
{%- set ns.found = true -%}
|
37 |
+
{%- endif -%}
|
38 |
+
{%- endfor -%}
|
39 |
+
{%- for message in messages %}
|
40 |
+
{%- if message['role'] == 'system' -%}
|
41 |
+
{{- '<|im_start|>system\n' + message['content'].rstrip() + '<|im_end|>\n' -}}
|
42 |
+
{%- else -%}
|
43 |
+
{%- if message['role'] == 'user' -%}
|
44 |
+
{{-'<|im_start|>human\n' + message['content'].rstrip() + '<|im_end|>\n'-}}
|
45 |
+
{%- else -%}
|
46 |
+
{{-'<|im_start|>assistant\n' + message['content'] + '<|im_end|>\n' -}}
|
47 |
+
{%- endif -%}
|
48 |
+
{%- endif -%}
|
49 |
+
{%- endfor -%}
|
50 |
+
{%- if add_generation_prompt -%}
|
51 |
+
{{-'<|im_start|>assistant\n'-}}
|
52 |
+
{%- endif -%}
|
53 |
+
```
|
54 |
+
|
55 |
+
- **Developed by:** gronkomatic
|
56 |
+
- **Funded by:** gronkomatic
|
57 |
+
- **Shared by:** gronkomatic
|
58 |
+
- **Model type:** Mistral
|
59 |
+
- **Language(s) (NLP):** English, maybe others I dunno
|
60 |
+
- **License:** OpenRAIL, IDGAF
|
61 |
+
|
62 |
+
### Model Sources
|
63 |
+
|
64 |
+
Exclusively available right here on HuggingFace!
|
65 |
+
|
66 |
+
- **Repository:** https://huggingface.co/neoncortex/mini-mistral-openhermes-2.5-chatml-test
|
67 |
+
- **Paper:** LoL
|
68 |
+
- **Demo:** Just download it in Oobabooga and use the modified chatML template above. Maybe I'll throw together a Space or something.
|
69 |
|
70 |
## Uses
|
71 |
|
72 |
+
If you wanna have a laugh at how bad it is then go ahead, but I wouldn't expect much from it.
|
73 |
|
74 |
### Direct Use
|
75 |
|
|
|
85 |
|
86 |
### Out-of-Scope Use
|
87 |
|
88 |
+
This model won't work well for pretty much everything, probably.
|
89 |
|
90 |
[More Information Needed]
|
91 |
|
|
|
119 |
|
120 |
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
|
121 |
|
122 |
+
#### Preprocessing
|
|
|
|
|
123 |
|
124 |
+
I took the OpenHermes 2.5 dataset formatted it with ChatML.
|
125 |
|
126 |
#### Training Hyperparameters
|
127 |
|
128 |
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
|
129 |
|
130 |
+
#### Speeds, Sizes, Times
|
131 |
+
|
132 |
+
epochs: 9
|
133 |
+
steps: 140976
|
134 |
+
batches per device: 6
|
135 |
+
1.04it/s
|
136 |
|
137 |
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
|
138 |
|
|
|
140 |
|
141 |
## Evaluation
|
142 |
|
143 |
+
I tried to run evals but the eval suite just laughed at me.
|
|
|
|
|
144 |
|
145 |
+
## Model Examination
|
146 |
|
147 |
+
Don't be rude.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
148 |
|
149 |
## Environmental Impact
|
150 |
|
|
|
152 |
|
153 |
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
|
154 |
|
155 |
+
- **Hardware Type:** I already told you. Try and keep up.
|
156 |
+
- **Hours used:** ~45 x 2 I guess.
|
157 |
+
- **Cloud Provider:** gronkomatic
|
158 |
+
- **Compute Region:** myob
|
159 |
+
- **Carbon Emitted:** Probably
|
160 |
|
161 |
+
## Technical Specifications
|
162 |
|
163 |
### Model Architecture and Objective
|
164 |
|
|
|
166 |
|
167 |
### Compute Infrastructure
|
168 |
|
169 |
+
I trained it on my PC with no side on it because I like to watch the GPUs do their work.
|
170 |
|
171 |
#### Hardware
|
172 |
|
173 |
+
2 x Nvidia RTX 3060 12GB
|
174 |
|
175 |
#### Software
|
176 |
|
177 |
+
The wonderful free stuff at HuggingFace (https://huggingface.co)[https://huggingface.co]: transformers, datasets, trl
|
|
|
|
|
|
|
|
|
178 |
|
179 |
+
## Glossary
|
180 |
|
181 |
+
IDGAF - I don't give a fuck
|
182 |
|
183 |
+
## More Information
|
184 |
|
185 |
[More Information Needed]
|
186 |
|
187 |
+
## Model Card Authors
|
|
|
|
|
|
|
|
|
188 |
|
189 |
+
gronkomatic, unless you're offended by something, in which case it was hacked by hackers.
|
|
|
|
|
|
|
|
|
|
|
|
|
190 |
|
191 |
## Model Card Contact
|
192 |
|
193 |
+
If you want to send me insults come find me on Reddit I guess u/gronkomatic.
|