Updated model

app.py

@@ -1,8 +1,10 @@
 # -*- coding: utf-8 -*-
 
+import os
+os.environ["TOKENIZERS_PARALLELISM"] = "false"
+
 import argparse
 import re
-import os
 
 import streamlit as st
 import random
@@ -11,8 +13,6 @@ import torch
 from transformers import AutoTokenizer, AutoModelForCausalLM
 import tokenizers
 
-#os.environ["TOKENIZERS_PARALLELISM"] = "false"
-
 random.seed(None)
 suggested_text_list = ['השד הופיע מול','קאלי שלפה את','פעם אחת לפני שנים רבות', 'הארי פוטר חייך חיוך נבוך', 'ואז הפרתי את כל כללי הטקס כש']
 
@@ -24,7 +24,9 @@ def load_model(model_name):
 
 def extend(input_text, max_size=20, top_k=50, top_p=0.95, temperature=0.7):
     if len(input_text) == 0:
-        input_text = ""
+        input_text = "<|startoftext|>"
+    else:
+        input_text = "<|startoftext|>" + input_text
 
     encoded_prompt = tokenizer.encode(
     input_text, add_special_tokens=False, return_tensors="pt")
@@ -83,7 +85,7 @@ if __name__ == "__main__":
     model, tokenizer = load_model("./model")
     
     stop_token = "<|endoftext|>"
-    new_lines = "\n\n\n"
+    new_lines = "<|pad|>"
 
     np.random.seed(None)
     random_seed = np.random.randint(10000,size=1)    

model/config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "./FantasyChildrenScifi-hebrew-gpt_neo-small/model",
+  "_name_or_path": "Norod78/hebrew-gpt_neo-small",
   "activation_function": "gelu_new",
   "architectures": [
     "GPTNeoForCausalLM"

model/pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:25e83f166977308069becae45ac59d48a2c08c0de8b3135a9acb63455fc0aec9
+oid sha256:25bdb066c638e0f8e2da703f4ecd9223448fcf282fd46e354a47e817656af8e5
 size 551197393

model/tokenizer_config.json

@@ -7,7 +7,7 @@
   "errors": "replace",
   "full_tokenizer_file": null,
   "max_len": 1024,
-  "name_or_path": "./FantasyChildrenScifi-hebrew-gpt_neo-small/model",
+  "name_or_path": "Norod78/hebrew-gpt_neo-small",
   "pad_token": "<|pad|>",
   "special_tokens_map_file": "special_tokens_map.json",
   "tokenizer_class": "GPT2Tokenizer",
@@ -18,5 +18,6 @@
     "normalized": true,
     "rstrip": false,
     "single_word": false
-  }
+  },
+  "unknown_token": "<|unknown|>"
 }