Archives/gpt4all
2
0
Fork 0
mirror of https://github.com/nomic-ai/gpt4all synced 2024-11-08 07:10:32 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix: tokenization error

Browse Source
This commit is contained in:
Zach Nussbaum 2023-04-08 20:33:51 +00:00
parent df29c62521
commit c0a9065032
1 changed files with 5 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
data.py
View File

@ -9,10 +9,6 @@ from transformers import DefaultDataCollator
def tokenize_inputs(config, tokenizer, examples): def tokenize_inputs(config, tokenizer, examples):
max_length = config["max_length"] max_length = config["max_length"]
# ignore bos
newline_tokens = tokenizer("\n", return_tensors="pt")["input_ids"][0]
if newline_tokens[0] == tokenizer.bos_token_id:
newline_tokens = newline_tokens[1:]
# hacky backward compatible # hacky backward compatible
different_eos = tokenizer.eos_token != "</s>" different_eos = tokenizer.eos_token != "</s>"
@ -22,7 +18,7 @@ def tokenize_inputs(config, tokenizer, examples):
if response.count("</s>") > 0: if response.count("</s>") > 0:
response = response.replace("</s>", tokenizer.eos_token) response = response.replace("</s>", tokenizer.eos_token)
prompt_len = len(tokenizer(prompt, return_tensors="pt")["input_ids"][0]) prompt_len = len(tokenizer(prompt + "\n", return_tensors="pt")["input_ids"][0])
# hack if our prompt is super long # hack if our prompt is super long
# we need to include some labels so we arbitrarily trunacate at max_length // 2 # we need to include some labels so we arbitrarily trunacate at max_length // 2
@ -33,7 +29,7 @@ def tokenize_inputs(config, tokenizer, examples):
new_len = min(max_length // 2, len(prompt) // 2) new_len = min(max_length // 2, len(prompt) // 2)
prompt = prompt[:new_len] prompt = prompt[:new_len]
# get new prompt length # get new prompt length
prompt_len = tokenizer(prompt, return_tensors="pt", max_length=max_length // 2, truncation=True).input_ids.ne(tokenizer.pad_token_id).sum().item() prompt_len = tokenizer(prompt + "\n", return_tensors="pt", max_length=max_length // 2, truncation=True).input_ids.ne(tokenizer.pad_token_id).sum().item()
assert prompt_len <= max_length // 2, f"prompt length {prompt_len} exceeds max length {max_length}" assert prompt_len <= max_length // 2, f"prompt length {prompt_len} exceeds max length {max_length}"
@ -41,11 +37,13 @@ def tokenize_inputs(config, tokenizer, examples):
truncation=True, max_length=max_length, return_tensors="pt")["input_ids"].squeeze() truncation=True, max_length=max_length, return_tensors="pt")["input_ids"].squeeze()
labels = input_tokens.clone() labels = input_tokens.clone()
labels[:prompt_len + len(newline_tokens)] = -100 labels[:prompt_len] = -100
if len(labels) < max_length: if len(labels) < max_length:
# pad to max_length with -100 # pad to max_length with -100
labels = torch.cat([labels, torch.full((max_length - len(labels),), -100)]) labels = torch.cat([labels, torch.full((max_length - len(labels),), -100)])
assert (labels == -100).sum() < len(labels), f"Labels are all -100, something wrong. prompt length {prompt_len} exceeds max length {max_length}"
if (labels == -100).sum() == len(labels) - 1: if (labels == -100).sum() == len(labels) - 1:
print(prompt) print(prompt)
print(response) print(response)