Update README.md
Browse files
README.md
CHANGED
@@ -7,41 +7,34 @@ model-index:
|
|
7 |
results: []
|
8 |
---
|
9 |
|
10 |
-
|
11 |
-
|
12 |
|
13 |
-
|
|
|
|
|
|
|
|
|
|
|
14 |
|
15 |
-
This
|
16 |
-
It achieves the following results on the evaluation set:
|
17 |
|
|
|
|
|
|
|
|
|
18 |
|
19 |
-
|
|
|
|
|
|
|
|
|
20 |
|
21 |
-
|
22 |
|
23 |
-
|
24 |
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
More information needed
|
30 |
-
|
31 |
-
## Training procedure
|
32 |
-
|
33 |
-
### Training hyperparameters
|
34 |
-
|
35 |
-
The following hyperparameters were used during training:
|
36 |
-
- optimizer: None
|
37 |
-
- training_precision: float32
|
38 |
-
|
39 |
-
### Training results
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
### Framework versions
|
44 |
-
|
45 |
-
- Transformers 4.42.0.dev0
|
46 |
-
- TensorFlow 2.11.0
|
47 |
-
- Tokenizers 0.19.1
|
|
|
7 |
results: []
|
8 |
---
|
9 |
|
10 |
+
Turns out that [Mistral-7B-Instruct-v0.3](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3) only have safetensors. This repo
|
11 |
+
is created to have the `.bin` files of the model.
|
12 |
|
13 |
+
This repo is created by:
|
14 |
+
```py
|
15 |
+
model_id = "mistralai/Mistral-7B-Instruct-v0.3"
|
16 |
+
model = AutoModelForCausalLM.from_pretrained(model_id)
|
17 |
+
model.push_to_hub("ariG23498/Mistral-7B-Instruct-v0.3", safe_serialization=False)
|
18 |
+
```
|
19 |
|
20 |
+
This is due to the fact that the TensorFlow port cannot use safetensors and need bin files.
|
|
|
21 |
|
22 |
+
You can use this model with TF like so:
|
23 |
+
```py
|
24 |
+
model_tf = TFAutoModelForCausalLM.from_pretrained("ariG23498/Mistral-7B-Instruct-v0.3", from_pt=True)
|
25 |
+
tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.3")
|
26 |
|
27 |
+
prompt = "My favourite condiment is"
|
28 |
+
model_inputs = tokenizer([prompt], return_tensors="tf")
|
29 |
+
generated_ids = model_tf.generate(**model_inputs, max_new_tokens=100, do_sample=True)
|
30 |
+
tokenizer.batch_decode(generated_ids)[0]
|
31 |
+
```
|
32 |
|
33 |
+
As soon as the safetensors and TensorFlow issue is sorted one can ditch this repository and use the official repository!
|
34 |
|
35 |
+
Update:
|
36 |
|
37 |
+
I have uploaded the `.h5` models as well. You can now use the following and make the entire code work!
|
38 |
+
```py
|
39 |
+
model_tf = TFAutoModelForCausalLM.from_pretrained("ariG23498/Mistral-7B-Instruct-v0.3")
|
40 |
+
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|