Update README.md
Browse files
README.md
CHANGED
@@ -47,6 +47,7 @@ You are a helpful, respectful and honest assistant. Always answer as helpfully a
|
|
47 |
|
48 |
{prompt} [/INST] {model_reply} [INST] {prompt} [/INST]
|
49 |
```
|
|
|
50 |
|
51 |
### Example usage
|
52 |
An example question you can ask:
|
@@ -89,7 +90,6 @@ Don't be evil.
|
|
89 |
|
90 |
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
|
91 |
- It's not quite perfect Dutch yet, but a very promising start.
|
92 |
-
- The model rarely generates EOS tokens and goes on a ramble. A quick and dirty way to mitigate this is to system prompt it to generate an `</s>` after 5 sentences.
|
93 |
|
94 |
### Recommendations
|
95 |
|
@@ -98,7 +98,8 @@ Don't be evil.
|
|
98 |
|
99 |
## How to Get Started with the Model
|
100 |
|
101 |
-
If you already have a pipeline running llama 2 7B chat in huggingface format, just call this one instead.
|
|
|
102 |
|
103 |
## Training Details
|
104 |
|
@@ -124,6 +125,7 @@ If you already have a pipeline running llama 2 7B chat in huggingface format, ju
|
|
124 |
#### Speeds, Sizes, Times [optional]
|
125 |
|
126 |
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
|
|
|
127 |
|
128 |
[More Information Needed]
|
129 |
|
@@ -169,15 +171,7 @@ If you already have a pipeline running llama 2 7B chat in huggingface format, ju
|
|
169 |
|
170 |
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
|
171 |
|
172 |
-
|
173 |
-
|
174 |
-
- **Hardware Type:** [More Information Needed]
|
175 |
-
- **Hours used:** [More Information Needed]
|
176 |
-
- **Cloud Provider:** [More Information Needed]
|
177 |
-
- **Compute Region:** [More Information Needed]
|
178 |
-
- **Carbon Emitted:** [More Information Needed]
|
179 |
-
|
180 |
-
## Technical Specifications [optional]
|
181 |
|
182 |
### Model Architecture and Objective
|
183 |
|
|
|
47 |
|
48 |
{prompt} [/INST] {model_reply} [INST] {prompt} [/INST]
|
49 |
```
|
50 |
+
**N.B.** Set your pad_token_id=18610 in your generator, otherwise it returns gibberish.
|
51 |
|
52 |
### Example usage
|
53 |
An example question you can ask:
|
|
|
90 |
|
91 |
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
|
92 |
- It's not quite perfect Dutch yet, but a very promising start.
|
|
|
93 |
|
94 |
### Recommendations
|
95 |
|
|
|
98 |
|
99 |
## How to Get Started with the Model
|
100 |
|
101 |
+
If you already have a pipeline running llama 2 7B chat in huggingface format, just call this one instead.
|
102 |
+
**N.B.** Set pad_token_id=18610 in your generator, otherwise it returns gibberish.
|
103 |
|
104 |
## Training Details
|
105 |
|
|
|
125 |
#### Speeds, Sizes, Times [optional]
|
126 |
|
127 |
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
|
128 |
+
We reached 32 tokens/second on a V100S without trying anything fancy.
|
129 |
|
130 |
[More Information Needed]
|
131 |
|
|
|
171 |
|
172 |
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
|
173 |
|
174 |
+
Yes.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
175 |
|
176 |
### Model Architecture and Objective
|
177 |
|