Update README.md
Browse files
README.md
CHANGED
|
@@ -1,7 +1,6 @@
|
|
| 1 |
---
|
| 2 |
language:
|
| 3 |
- tr
|
| 4 |
-
- en
|
| 5 |
tags:
|
| 6 |
- text-generation
|
| 7 |
- custom-architecture
|
|
@@ -154,6 +153,9 @@ def generate_text(prompt, max_new_tokens=60, temperature=0.5, top_k=20):
|
|
| 154 |
full_prompt = f"Kullanıcı: {prompt}\nModel: "
|
| 155 |
input_ids = tokenizer.encode(full_prompt)
|
| 156 |
idx = torch.tensor(input_ids, dtype=torch.long, device=device).unsqueeze(0)
|
|
|
|
|
|
|
|
|
|
| 157 |
|
| 158 |
for _ in range(max_new_tokens):
|
| 159 |
idx_cond = idx[:, -config.block_size:]
|
|
@@ -165,17 +167,33 @@ def generate_text(prompt, max_new_tokens=60, temperature=0.5, top_k=20):
|
|
| 165 |
logits[logits < v[:, [-1]]] = -float('Inf')
|
| 166 |
probs = F.softmax(logits, dim=-1)
|
| 167 |
idx_next = torch.multinomial(probs, num_samples=1)
|
| 168 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 169 |
idx = torch.cat((idx, idx_next), dim=1)
|
| 170 |
|
| 171 |
-
|
| 172 |
-
|
| 173 |
-
return response
|
| 174 |
|
| 175 |
# --- ÖRNEK KULLANIM / EXAMPLE USAGE ---
|
| 176 |
soru = "Nasılsın?"
|
| 177 |
cevap = generate_text(soru)
|
| 178 |
print(f"Soru: {soru}\nCevap: {cevap}")
|
|
|
|
|
|
|
|
|
|
|
|
|
| 179 |
```
|
| 180 |
|
| 181 |
---
|
|
|
|
| 1 |
---
|
| 2 |
language:
|
| 3 |
- tr
|
|
|
|
| 4 |
tags:
|
| 5 |
- text-generation
|
| 6 |
- custom-architecture
|
|
|
|
| 153 |
full_prompt = f"Kullanıcı: {prompt}\nModel: "
|
| 154 |
input_ids = tokenizer.encode(full_prompt)
|
| 155 |
idx = torch.tensor(input_ids, dtype=torch.long, device=device).unsqueeze(0)
|
| 156 |
+
|
| 157 |
+
|
| 158 |
+
generated_ids = []
|
| 159 |
|
| 160 |
for _ in range(max_new_tokens):
|
| 161 |
idx_cond = idx[:, -config.block_size:]
|
|
|
|
| 167 |
logits[logits < v[:, [-1]]] = -float('Inf')
|
| 168 |
probs = F.softmax(logits, dim=-1)
|
| 169 |
idx_next = torch.multinomial(probs, num_samples=1)
|
| 170 |
+
|
| 171 |
+
|
| 172 |
+
generated_ids.append(idx_next.item())
|
| 173 |
+
|
| 174 |
+
|
| 175 |
+
decoded_so_far = tokenizer.decode(generated_ids)
|
| 176 |
+
if "Kullanıcı:" in decoded_so_far or "Model:" in decoded_so_far:
|
| 177 |
+
|
| 178 |
+
generated_ids = generated_ids[:-1]
|
| 179 |
+
break
|
| 180 |
+
|
| 181 |
+
if idx_next.item() == tokenizer.eos_id():
|
| 182 |
+
break
|
| 183 |
+
|
| 184 |
idx = torch.cat((idx, idx_next), dim=1)
|
| 185 |
|
| 186 |
+
response = tokenizer.decode(generated_ids)
|
| 187 |
+
return response.strip()
|
|
|
|
| 188 |
|
| 189 |
# --- ÖRNEK KULLANIM / EXAMPLE USAGE ---
|
| 190 |
soru = "Nasılsın?"
|
| 191 |
cevap = generate_text(soru)
|
| 192 |
print(f"Soru: {soru}\nCevap: {cevap}")
|
| 193 |
+
|
| 194 |
+
soru = "En sevdiğin renk ne?"
|
| 195 |
+
cevap = generate_text(soru)
|
| 196 |
+
print(f"Soru: {soru}\nCevap: {cevap}")
|
| 197 |
```
|
| 198 |
|
| 199 |
---
|