Update README.md
Browse files
README.md
CHANGED
@@ -59,6 +59,39 @@ git checkout v1.11.0
|
|
59 |
python megatron_gpt_eval.py gpt_model_file=nemo_gpt5B_fp16.nemo server=True tensor_model_parallel_size=1 trainer.devices=1
|
60 |
```
|
61 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
62 |
|
63 |
## Training Data
|
64 |
|
|
|
59 |
python megatron_gpt_eval.py gpt_model_file=nemo_gpt5B_fp16.nemo server=True tensor_model_parallel_size=1 trainer.devices=1
|
60 |
```
|
61 |
|
62 |
+
### Step 3: Send prompts to you model!
|
63 |
+
```python
|
64 |
+
import json
|
65 |
+
import requests
|
66 |
+
|
67 |
+
port_num = 5555
|
68 |
+
headers = {"Content-Type": "application/json"}
|
69 |
+
|
70 |
+
def request_data(data):
|
71 |
+
resp = requests.put('http://localhost:{}/generate'.format(port_num),
|
72 |
+
data=json.dumps(data),
|
73 |
+
headers=headers)
|
74 |
+
sentences = resp.json()['sentences']
|
75 |
+
return sentences
|
76 |
+
|
77 |
+
|
78 |
+
data = {
|
79 |
+
"sentences": ["Tell me an interesting fact about space travel."]*1,
|
80 |
+
"tokens_to_generate": 50,
|
81 |
+
"temperature": 1.0,
|
82 |
+
"add_BOS": True,
|
83 |
+
"top_k": 0,
|
84 |
+
"top_p": 0.9,
|
85 |
+
"greedy": False,
|
86 |
+
"all_probs": False,
|
87 |
+
"repetition_penalty": 1.2,
|
88 |
+
"min_tokens_to_generate": 2,
|
89 |
+
}
|
90 |
+
|
91 |
+
sentences = request_data(data)
|
92 |
+
print(sentences)
|
93 |
+
```
|
94 |
+
|
95 |
|
96 |
## Training Data
|
97 |
|