jacobrenn commited on
Commit
39d71b1
·
1 Parent(s): e4ac526

uploading model

Browse files

Signed-off-by: Jacob Renn <jacob.renn@squared.ai>

added_tokens.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "### End": 50257,
3
+ "### Instruction:": 50258,
4
+ "### Response:\n": 50259
5
+ }
config.json ADDED
@@ -0,0 +1,54 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "/dbfs/FileStore/dais-model-large",
3
+ "activation_function": "gelu_new",
4
+ "architectures": [
5
+ "GPT2LMHeadModel"
6
+ ],
7
+ "custom_pipelines": {
8
+ "text-generation": {
9
+ "impl": "instruct_pipeline.InstructionTextGenerationPipeline",
10
+ "pt": "AutoModelForCausalLM",
11
+ "tf": "TFAutoModelForCausalLM"
12
+ }
13
+ },
14
+ "attn_pdrop": 0.1,
15
+ "bos_token_id": 50256,
16
+ "custom_pipelines": {
17
+ "text-generation": {
18
+ "impl": "instruct_pipeline.InstructionTextGenerationPipeline",
19
+ "pt": "AutoModelForCausalLM",
20
+ "tf": "TFAutoModelForCausalLM"
21
+ }
22
+ },
23
+ "embd_pdrop": 0.1,
24
+ "eos_token_id": 50256,
25
+ "initializer_range": 0.02,
26
+ "layer_norm_epsilon": 1e-05,
27
+ "model_type": "gpt2",
28
+ "n_ctx": 1024,
29
+ "n_embd": 1600,
30
+ "n_head": 25,
31
+ "n_inner": null,
32
+ "n_layer": 48,
33
+ "n_positions": 1024,
34
+ "output_past": true,
35
+ "reorder_and_upcast_attn": false,
36
+ "resid_pdrop": 0.1,
37
+ "scale_attn_by_inverse_layer_idx": false,
38
+ "scale_attn_weights": true,
39
+ "summary_activation": null,
40
+ "summary_first_dropout": 0.1,
41
+ "summary_proj_to_labels": true,
42
+ "summary_type": "cls_index",
43
+ "summary_use_proj": true,
44
+ "task_specific_params": {
45
+ "text-generation": {
46
+ "do_sample": true,
47
+ "max_length": 50
48
+ }
49
+ },
50
+ "torch_dtype": "float16",
51
+ "transformers_version": "4.25.1",
52
+ "use_cache": false,
53
+ "vocab_size": 50260
54
+ }
config.json~ ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "/dbfs/FileStore/dais-model-large",
3
+ "activation_function": "gelu_new",
4
+ "architectures": [
5
+ "GPT2LMHeadModel"
6
+ ],
7
+ "attn_pdrop": 0.1,
8
+ "bos_token_id": 50256,
9
+ "custom_pipelines": {
10
+ "text-generation": {
11
+ "impl": "instruct_pipeline.InstructionTextGenerationPipeline",
12
+ "pt": "AutoModelForCausalLM",
13
+ "tf": "TFAutoModelForCausalLM"
14
+ }
15
+ },
16
+ "embd_pdrop": 0.1,
17
+ "eos_token_id": 50256,
18
+ "initializer_range": 0.02,
19
+ "layer_norm_epsilon": 1e-05,
20
+ "model_type": "gpt2",
21
+ "n_ctx": 1024,
22
+ "n_embd": 1600,
23
+ "n_head": 25,
24
+ "n_inner": null,
25
+ "n_layer": 48,
26
+ "n_positions": 1024,
27
+ "output_past": true,
28
+ "reorder_and_upcast_attn": false,
29
+ "resid_pdrop": 0.1,
30
+ "scale_attn_by_inverse_layer_idx": false,
31
+ "scale_attn_weights": true,
32
+ "summary_activation": null,
33
+ "summary_first_dropout": 0.1,
34
+ "summary_proj_to_labels": true,
35
+ "summary_type": "cls_index",
36
+ "summary_use_proj": true,
37
+ "task_specific_params": {
38
+ "text-generation": {
39
+ "do_sample": true,
40
+ "max_length": 50
41
+ }
42
+ },
43
+ "torch_dtype": "float16",
44
+ "transformers_version": "4.25.1",
45
+ "use_cache": false,
46
+ "vocab_size": 50260
47
+ }
instruct_pipeline.py ADDED
@@ -0,0 +1,160 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import re
2
+
3
+ import numpy as np
4
+ from transformers import Pipeline, PreTrainedTokenizer
5
+
6
+
7
+ INSTRUCTION_KEY = "### Instruction:"
8
+ RESPONSE_KEY = "### Response:"
9
+ END_KEY = "### End"
10
+ INTRO_BLURB = (
11
+ "Below is an instruction that describes a task, along with any additional context. Write a response that appropriately completes the request."
12
+ )
13
+
14
+ # This is the prompt that is used for generating responses using an already trained model. It ends with the response
15
+ # key, where the job of the model is to provide the completion that follows it (i.e. the response itself).
16
+ PROMPT_FOR_GENERATION_FORMAT = """{intro}
17
+
18
+ {instruction_key}
19
+ {instruction}
20
+
21
+ {response_key}
22
+ """.format(
23
+ intro=INTRO_BLURB,
24
+ instruction_key=INSTRUCTION_KEY,
25
+ instruction="{instruction}",
26
+ response_key=RESPONSE_KEY,
27
+ )
28
+
29
+
30
+ def get_special_token_id(tokenizer: PreTrainedTokenizer, key: str) -> int:
31
+ """Gets the token ID for a given string that has been added to the tokenizer as a special token.
32
+
33
+ When training, we configure the tokenizer so that the sequences like "### Instruction:" and "### End" are
34
+ treated specially and converted to a single, new token. This retrieves the token ID each of these keys map to.
35
+
36
+ Args:
37
+ tokenizer (PreTrainedTokenizer): the tokenizer
38
+ key (str): the key to convert to a single token
39
+
40
+ Raises:
41
+ RuntimeError: if more than one ID was generated
42
+
43
+ Returns:
44
+ int: the token ID for the given key
45
+ """
46
+ token_ids = tokenizer.encode(key)
47
+ if len(token_ids) > 1:
48
+ raise ValueError(f"Expected only a single token for '{key}' but found {token_ids}")
49
+ return token_ids[0]
50
+
51
+
52
+ class InstructionTextGenerationPipeline(Pipeline):
53
+ def __init__(
54
+ self, *args, do_sample: bool = True, max_new_tokens: int = 256, top_p: float = 0.92, top_k: int = 0, **kwargs
55
+ ):
56
+ super().__init__(*args, do_sample=do_sample, max_new_tokens=max_new_tokens, top_p=top_p, top_k=top_k, **kwargs)
57
+
58
+ def _sanitize_parameters(self, return_instruction_text=False, **generate_kwargs):
59
+ preprocess_params = {}
60
+
61
+ # newer versions of the tokenizer configure the response key as a special token. newer versions still may
62
+ # append a newline to yield a single token. find whatever token is configured for the response key.
63
+ tokenizer_response_key = next(
64
+ (token for token in self.tokenizer.additional_special_tokens if token.startswith(RESPONSE_KEY)), None
65
+ )
66
+
67
+ response_key_token_id = None
68
+ end_key_token_id = None
69
+ if tokenizer_response_key:
70
+ try:
71
+ response_key_token_id = get_special_token_id(self.tokenizer, tokenizer_response_key)
72
+ end_key_token_id = get_special_token_id(self.tokenizer, END_KEY)
73
+
74
+ # Ensure generation stops once it generates "### End"
75
+ generate_kwargs["eos_token_id"] = end_key_token_id
76
+ except ValueError:
77
+ pass
78
+
79
+ forward_params = generate_kwargs
80
+ postprocess_params = {
81
+ "response_key_token_id": response_key_token_id,
82
+ "end_key_token_id": end_key_token_id,
83
+ "return_instruction_text": return_instruction_text,
84
+ }
85
+
86
+ return preprocess_params, forward_params, postprocess_params
87
+
88
+ def preprocess(self, instruction_text, **generate_kwargs):
89
+ prompt_text = PROMPT_FOR_GENERATION_FORMAT.format(instruction=instruction_text)
90
+ inputs = self.tokenizer(
91
+ prompt_text,
92
+ return_tensors="pt",
93
+ )
94
+ inputs["prompt_text"] = prompt_text
95
+ inputs["instruction_text"] = instruction_text
96
+ return inputs
97
+
98
+ def _forward(self, model_inputs, **generate_kwargs):
99
+ input_ids = model_inputs["input_ids"]
100
+ attention_mask = model_inputs.get("attention_mask", None)
101
+ generated_sequence = self.model.generate(
102
+ input_ids=input_ids.to(self.model.device),
103
+ attention_mask=attention_mask,
104
+ pad_token_id=self.tokenizer.pad_token_id,
105
+ **generate_kwargs,
106
+ )[0].cpu()
107
+ instruction_text = model_inputs.pop("instruction_text")
108
+ return {"generated_sequence": generated_sequence, "input_ids": input_ids, "instruction_text": instruction_text}
109
+
110
+ def postprocess(self, model_outputs, response_key_token_id, end_key_token_id, return_instruction_text):
111
+ sequence = model_outputs["generated_sequence"]
112
+ instruction_text = model_outputs["instruction_text"]
113
+
114
+ # The response will be set to this variable if we can identify it.
115
+ decoded = None
116
+
117
+ # If we have token IDs for the response and end, then we can find the tokens and only decode between them.
118
+ if response_key_token_id and end_key_token_id:
119
+ # Find where "### Response:" is first found in the generated tokens. Considering this is part of the
120
+ # prompt, we should definitely find it. We will return the tokens found after this token.
121
+ response_pos = None
122
+ response_positions = np.where(sequence == response_key_token_id)[0]
123
+ if len(response_positions) == 0:
124
+ pass
125
+ else:
126
+ response_pos = response_positions[0]
127
+
128
+ if response_pos:
129
+ # Next find where "### End" is located. The model has been trained to end its responses with this
130
+ # sequence (or actually, the token ID it maps to, since it is a special token). We may not find
131
+ # this token, as the response could be truncated. If we don't find it then just return everything
132
+ # to the end. Note that even though we set eos_token_id, we still see the this token at the end.
133
+ end_pos = None
134
+ end_positions = np.where(sequence == end_key_token_id)[0]
135
+ if len(end_positions) > 0:
136
+ end_pos = end_positions[0]
137
+
138
+ decoded = self.tokenizer.decode(sequence[response_pos + 1 : end_pos]).strip()
139
+ else:
140
+ # Otherwise we'll decode everything and use a regex to find the response and end.
141
+
142
+ fully_decoded = self.tokenizer.decode(sequence)
143
+
144
+ # The response appears after "### Response:". The model has been trained to append "### End" at the
145
+ # end.
146
+ m = re.search(r"#+\s*Response:\s*(.+?)#+\s*End", fully_decoded, flags=re.DOTALL)
147
+
148
+ if m:
149
+ decoded = m.group(1).strip()
150
+ else:
151
+ # The model might not generate the "### End" sequence before reaching the max tokens. In this case,
152
+ # return everything after "### Response:".
153
+ m = re.search(r"#+\s*Response:\s*(.+)", fully_decoded, flags=re.DOTALL)
154
+ if m:
155
+ decoded = m.group(1).strip()
156
+
157
+ if return_instruction_text:
158
+ return {"instruction_text": instruction_text, "generated_text": decoded}
159
+
160
+ return decoded
merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:30a271cb3f0d38bbc53d2e86f933f3af9502818308bc6846707cdf7070966db4
3
+ size 3165775133
special_tokens_map.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "### End",
4
+ "### Instruction:",
5
+ "### Response:\n"
6
+ ],
7
+ "bos_token": "<|endoftext|>",
8
+ "eos_token": "<|endoftext|>",
9
+ "pad_token": "<|endoftext|>",
10
+ "unk_token": "<|endoftext|>"
11
+ }
tokenizer_config.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": false,
3
+ "add_prefix_space": false,
4
+ "bos_token": {
5
+ "__type": "AddedToken",
6
+ "content": "<|endoftext|>",
7
+ "lstrip": false,
8
+ "normalized": true,
9
+ "rstrip": false,
10
+ "single_word": false
11
+ },
12
+ "eos_token": {
13
+ "__type": "AddedToken",
14
+ "content": "<|endoftext|>",
15
+ "lstrip": false,
16
+ "normalized": true,
17
+ "rstrip": false,
18
+ "single_word": false
19
+ },
20
+ "errors": "replace",
21
+ "model_max_length": 1024,
22
+ "name_or_path": "/dbfs/FileStore/dais-model-large",
23
+ "pad_token": null,
24
+ "special_tokens_map_file": null,
25
+ "tokenizer_class": "GPT2Tokenizer",
26
+ "unk_token": {
27
+ "__type": "AddedToken",
28
+ "content": "<|endoftext|>",
29
+ "lstrip": false,
30
+ "normalized": true,
31
+ "rstrip": false,
32
+ "single_word": false
33
+ }
34
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b0b3f10c8b1c6eb7e6085ee8166b3eedc44b8b9255e92da9797f51d5648462c5
3
+ size 4603
vocab.json ADDED
The diff for this file is too large to render. See raw diff