#7 add simple api

scripts/promptgen.py

@@ -1,6 +1,7 @@
 import html
 import os
 import time
+from pydantic import BaseModel, Field
 
 import torch
 import transformers
@@ -18,12 +19,29 @@ class Model:
     model = None
     tokenizer = None
 
+class ApiResult(BaseModel):
+    items: list = Field(title="Generated prompts")
+
+class GenerateRequest(BaseModel):
+    prompt: str = Field(title="Beginning of the prompt")
+    sampling_mode: str = Field(title="The mode", default='Top K')
+    repetition_penalty: int = Field(title="Repetition penalty", default=1)
+    length_penalty: int = Field(title="Length penalty", default=1)
+    min_length: int = Field(title="Min length", default=20)
+    max_length: int = Field(title="Max length", default=150)
+    num_beams: int = Field(title="Number of beams", default=1)
+    top_k: int = Field(title="Top K", default=12)
+    top_p: float = Field(title="Top P", default=0.15)
+    temperature: float = Field(title="Temperature", default=1)
+    batch_count: int = Field(title="How many batch", default=1)
+    batch_size: int = Field(title="How many to generate", default=10)
 
 available_models = []
 current = Model()
 
 base_dir = scripts.basedir()
 models_dir = os.path.join(base_dir, "models")
+current_api_result = ApiResult(items = [])
 
 
 def device():
@@ -115,8 +133,10 @@ def generate(id_task, model_name, batch_count, batch_size, text, *args):
     markup = '<table><tbody>'
 
     index = 0
+    all_texts = []
     for i in range(batch_count):
         texts = generate_batch(input_ids, *args)
+        all_texts = all_texts + texts
         shared.state.nextjob()
         for generated_text in texts:
             index += 1
@@ -135,6 +155,7 @@ def generate(id_task, model_name, batch_count, batch_size, text, *args):
 """
 
     markup += '</tbody></table>'
+    current_api_result.items = all_texts
 
     return markup, ''
 
@@ -237,7 +258,16 @@ def on_unload():
     current.model = None
     current.tokenizer = None
 
+def api_requested(req: GenerateRequest):
+    generate('', available_models[0], req.batch_count, req.batch_size, req.prompt, req.min_length, req.max_length, req.num_beams, req.temperature, req.repetition_penalty, req.length_penalty, req.sampling_mode, req.top_k, req.top_p)
+    return current_api_result
+
+def register_api(demo, app):
+    if shared.cmd_opts.api:
+        app.add_api_route("/promptgen", api_requested, methods=["POST"], response_model=ApiResult)
+
 
 script_callbacks.on_ui_tabs(add_tab)
 script_callbacks.on_ui_settings(on_ui_settings)
 script_callbacks.on_script_unloaded(on_unload)
+script_callbacks.on_app_started(register_api)