Andrei Panferov commited on
Commit
d6e09e8
1 Parent(s): 2f56b8d

tokenizer and such

Browse files
README.md CHANGED
@@ -1,3 +1,57 @@
1
  ---
2
- license: bsd-2-clause
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ license: bsd-3-clause
3
  ---
4
+ # CodeGen (CodeGen-Multi 16B)
5
+
6
+ ## Model description
7
+
8
+ CodeGen is a family of autoregressive language models for **program synthesis** from the paper: [A Conversational Paradigm for Program Synthesis](https://arxiv.org/abs/2203.13474) by Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, Caiming Xiong. The models are originally released in [this repository](https://github.com/salesforce/CodeGen), under 3 pre-training data variants (`NL`, `Multi`, `Mono`) and 4 model size variants (`350M`, `2B`, `6B`, `16B`).
9
+
10
+ The checkpoint included in this repository is denoted as **CodeGen-Multi 16B** in the paper, where "Multi" means the model is initialized with *CodeGen-NL 16B* and further pre-trained on a dataset of multiple programming languages, and "16B" refers to the number of trainable parameters.
11
+
12
+ ## Training data
13
+
14
+ This checkpoint (CodeGen-Multi 16B) was firstly initialized with *CodeGen-NL 16B*, and then pre-trained on [BigQuery](https://console.cloud.google.com/marketplace/details/github/github-repos), a large-scale dataset of multiple programming languages from GitHub repositories. The data consists of 119.2B tokens and includes C, C++, Go, Java, JavaScript, and Python.
15
+
16
+ ## Training procedure
17
+
18
+ CodeGen was trained using cross-entropy loss to maximize the likelihood of sequential inputs.
19
+ The family of models are trained using multiple TPU-v4-512 by Google, leveraging data and model parallelism.
20
+ See Section 2.3 of the [paper](https://arxiv.org/abs/2203.13474) for more details.
21
+
22
+ ## Evaluation results
23
+
24
+ We evaluate our models on two code generation benchmark: HumanEval and MTPB. Please refer to the [paper](https://arxiv.org/abs/2203.13474) for more details.
25
+
26
+
27
+ ## Intended Use and Limitations
28
+
29
+ As an autoregressive language model, CodeGen is capable of extracting features from given natural language and programming language texts, and calculating the likelihood of them.
30
+ However, the model is intended for and best at **program synthesis**, that is, generating executable code given English prompts, where the prompts should be in the form of a comment string. The model can complete partially-generated code as well.
31
+
32
+ ## How to use
33
+
34
+ This model can be easily loaded using the `AutoModelForCausalLM` functionality:
35
+
36
+ ```python
37
+ from transformers import AutoTokenizer, AutoModelForCausalLM
38
+ tokenizer = AutoTokenizer.from_pretrained("Salesforce/codegen-16B-multi")
39
+ model = AutoModelForCausalLM.from_pretrained("Salesforce/codegen-16B-multi")
40
+
41
+ text = "def hello_world():"
42
+ input_ids = tokenizer(text, return_tensors="pt").input_ids
43
+
44
+ generated_ids = model.generate(input_ids, max_length=128)
45
+ print(tokenizer.decode(generated_ids[0], skip_special_tokens=True))
46
+ ```
47
+
48
+ ## BibTeX entry and citation info
49
+
50
+ ```bibtex
51
+ @article{Nijkamp2022ACP,
52
+ title={A Conversational Paradigm for Program Synthesis},
53
+ author={Nijkamp, Erik and Pang, Bo and Hayashi, Hiroaki and Tu, Lifu and Wang, Huan and Zhou, Yingbo and Savarese, Silvio and Xiong, Caiming},
54
+ journal={arXiv preprint},
55
+ year={2022}
56
+ }
57
+ ```
added_tokens.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {" ": 50280, " ": 50284, " ": 50262, " ": 50266, "\t\t\t\t\t\t\t": 50289, " ": 50264, " ": 50279, " ": 50281, "\t\t\t\t\t\t\t\t\t": 50287, " ": 50286, "\t\t\t": 50293, " ": 50261, " ": 50282, " ": 50283, " ": 50269, " ": 50273, " ": 50271, "\t\t\t\t\t\t\t\t": 50288, " ": 50285, " ": 50276, "\t\t\t\t\t\t": 50290, "\t\t\t\t\t": 50291, " ": 50263, " ": 50278, " ": 50258, " ": 50270, " ": 50259, " ": 50272, " ": 50274, " ": 50267, " ": 50268, "\t\t": 50294, " ": 50257, " ": 50277, "\t\t\t\t": 50292, " ": 50260, " ": 50265, " ": 50275}
config.json CHANGED
@@ -36,7 +36,7 @@
36
  "tie_word_embeddings": false,
37
  "tokenizer_class": "GPT2Tokenizer",
38
  "torch_dtype": "float16",
39
- "transformers_version": "4.28.0.dev0",
40
  "use_cache": true,
41
  "vocab_size": 51200
42
  }
 
36
  "tie_word_embeddings": false,
37
  "tokenizer_class": "GPT2Tokenizer",
38
  "torch_dtype": "float16",
39
+ "transformers_version": "4.21.0.dev0",
40
  "use_cache": true,
41
  "vocab_size": 51200
42
  }
merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
special_tokens_map.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"bos_token": "<|endoftext|>", "eos_token": "<|endoftext|>", "unk_token": "<|endoftext|>"}
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"unk_token": "<|endoftext|>", "bos_token": "<|endoftext|>", "eos_token": "<|endoftext|>", "add_prefix_space": false, "model_max_length": 2048, "special_tokens_map_file": null, "name_or_path": "gpt2", "tokenizer_class": "CodeGenTokenizer"}
vocab.json ADDED
The diff for this file is too large to render. See raw diff