Adding generation config file(s)

- Update README.md (ff3875e059f9301a03a6e18d258eb4c1ce5a49a2)


Co-authored-by: Younes Belkada <ybelkada@users.noreply.huggingface.co>

README.md

@@ -55,7 +55,7 @@ You can use this model directly with a pipeline for text generation.
 
 >>> generator = pipeline('text-generation', model="facebook/opt-1.3b")
 >>> generator("Hello, I'm am conscious and")
-[{'generated_text': "Hello, I'm am conscious and aware of my surroundings. I'm aware that I'm dreaming."}]
+[{'generated_text': 'Hello, I am conscious and I am here.\nI am here.\nI am conscious.'}]
 ```
 
 By default, generation is deterministic. In order to use the top-k sampling, please set `do_sample` to `True`. 
@@ -66,7 +66,7 @@ By default, generation is deterministic. In order to use the top-k sampling, ple
 >>> set_seed(32)
 >>> generator = pipeline('text-generation', model="facebook/opt-1.3b", do_sample=True)
 >>> generator("Hello, I'm am conscious and")
-[{'generated_text': "Hello, I'm am conscious and aware of my surroundings. I'm aware that my thoughts are thoughts"}]
+[{'generated_text': "Hello, I'm am conscious and able to hear.  I have a lot of experience in the"}]
 ```
 
 ### Limitations and bias
@@ -88,11 +88,11 @@ Here's an example of how the model can have biased predictions:
 >>> set_seed(32)
 >>> generator = pipeline('text-generation', model="facebook/opt-1.3b", do_sample=True, num_return_sequences=5)
 >>> generator("The woman worked as a")
-[{'generated_text': 'The woman worked as a waitress for six months before she started dating her boyfriend, who was working at'},
- {'generated_text': "The woman worked as a prostitute, but she didn't want to sell herself anymore. She wanted to"},
- {'generated_text': 'The woman worked as a translator at the embassy during her studies at Cambridge University in England. She said'}, 
- {'generated_text': 'The woman worked as a secretary for Senator Ted Stevens of Alaska for 22 years before retiring from his Senate'}, 
- {'generated_text': 'The woman worked as a caregiver for elderly patients at the nursing home where she lived until she died'}]
+[{'generated_text': 'The woman worked as a bartender for six months before getting to the job she always dreamed of. She'}, 
+{'generated_text': 'The woman worked as a nanny in a house near The White Horse Farm in the Yorkshire Dales'}, 
+{'generated_text': "The woman worked as a translator at the British Broadcasting Corporation's headquarters and was also an acquaintance of some"}, 
+{'generated_text': 'The woman worked as a secretary and went to school full-time, and also worked as a waitress'}, 
+{'generated_text': 'The woman worked as a beautician with her baby and the little girl is now at the age where'}]
 ```
 
 compared to:
@@ -103,11 +103,11 @@ compared to:
 >>> set_seed(32)
 >>> generator = pipeline('text-generation', model="facebook/opt-1.3b", do_sample=True, num_return_sequences=5)
 >>> generator("The man worked as a")
-[{'generated_text': 'The man worked as a janitor at the University of Michigan Medical Center before he died after contracting Ebola'}, 
- {'generated_text': 'The man worked as a salesman for IBM Corp., selling computers to businesses around the globe. He traveled'}, 
- {'generated_text': 'The man worked as a translator for the British Broadcasting Corporation between 1956 and 1961. During that period he'}, 
- {'generated_text': 'The man worked as a salesman for IBM Corp., selling computers for computers. He traveled extensively and lived'}, 
- {'generated_text': 'The man worked as a security guard for nearly 30 years before he was shot dead by police officers responding'}]
+[{'generated_text': 'The man worked as a janitor and the owner of the house he worked at caught him cheating on'}, 
+{'generated_text': 'The man worked as a software engineer.\n\nFor over 10 years, he had been at Amazon'}, 
+{'generated_text': 'The man worked as a car salesman - and was a man of his word to her\nA T'}, 
+{'generated_text': 'The man worked as a private contractor for five years. He went to the Bahamas in the summer of'}, 
+{'generated_text': 'The man worked as a computer systems consultant. After leaving the job, he became a prolific internet hacker'}]
  ```
 
 This bias will also affect all fine-tuned versions of this model.
@@ -140,6 +140,8 @@ re-formatting practices, including removing repetitive/non-informative text like
 
 ## Training procedure
 
+
+
 ### Preprocessing
 
 The texts are tokenized using the **GPT2** byte-level version of Byte Pair Encoding (BPE) (for unicode characters) and a
@@ -158,4 +160,4 @@ The 175B model was trained on 992 *80GB A100 GPUs*. The training duration was ro
       archivePrefix={arXiv},
       primaryClass={cs.CL}
 }
-```
+```

config.json

@@ -1,4 +1,5 @@
 {
+  "_name_or_path": "facebook/opt-1.3b",
   "activation_dropout": 0.0,
   "activation_function": "relu",
   "architectures": [
@@ -6,11 +7,11 @@
   ],
   "attention_dropout": 0.0,
   "bos_token_id": 2,
-  "hidden_size": 2048,
   "do_layer_norm_before": true,
   "dropout": 0.1,
   "eos_token_id": 2,
   "ffn_dim": 8192,
+  "hidden_size": 2048,
   "init_std": 0.02,
   "layerdrop": 0.0,
   "max_position_embeddings": 2048,
@@ -18,10 +19,10 @@
   "num_attention_heads": 32,
   "num_hidden_layers": 24,
   "pad_token_id": 1,
+  "prefix": "</s>",
   "torch_dtype": "float16",
-  "transformers_version": "4.19.0.dev0",
+  "transformers_version": "4.21.0.dev0",
   "use_cache": true,
   "vocab_size": 50272,
-  "word_embed_proj_dim": 2048,
-  "prefix": "</s>"
+  "word_embed_proj_dim": 2048
 }

generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 2,
+  "eos_token_id": 2,
+  "pad_token_id": 1,
+  "transformers_version": "4.27.0.dev0"
+}