Initial deploy

README.md

@@ -1,13 +1,66 @@
 ---
-title: Dynamic Transformers Api
-emoji: 📉
-colorFrom: purple
-colorTo: yellow
+title: Dynamic Transformers Pipeline API
+emoji: 🚀
+colorFrom: blue
+colorTo: purple
 sdk: gradio
-sdk_version: 6.22.0
-python_version: '3.13'
+sdk_version: 4.x
 app_file: app.py
 pinned: false
+license: apache-2.0
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Dynamic Transformers Pipeline API
+
+Pass any `transformers.pipeline()` config as JSON, run inference on ZeroGPU RTX Pro 6000 Blackwell.
+
+## API Usage
+
+```python
+from gradio_client import Client
+
+client = Client("your-username/your-space")
+result = client.predict(
+    pipeline_config={"task": "text-generation", "model": "HuggingFaceTB/SmolLM2-135M-Instruct"},
+    inputs="Hello world",
+    inference_kwargs={"max_new_tokens": 50},
+    api_name="/inference"
+)
+```
+
+
+---
+
+## 3. Hit deploy
+
+Upload → Space builds (~30-60s) → done.
+
+---
+
+## 4. Test it
+
+Replace `your-username/your-space` with your actual Space name:
+
+```python
+from gradio_client import Client
+
+client = Client("your-username/dynamic-transformers-api")
+
+# Text generation
+print(client.predict(
+    pipeline_config={"task": "text-generation", "model": "HuggingFaceTB/SmolLM2-135M-Instruct"},
+    inputs="The future of AI is",
+    inference_kwargs={"max_new_tokens": 50},
+    api_name="/inference"
+))
+
+# NER (swap pipeline on the fly, same Space)
+print(client.predict(
+    pipeline_config={"task": "ner", "model": "Jean-Baptiste/camembert-ner", "aggregation_strategy": "simple"},
+    inputs="Apple is looking at buying U.K. startup for $1 billion",
+    inference_kwargs={},
+    api_name="/inference"
+))
+```
+
+

app.py

@@ -0,0 +1,88 @@
+import gradio as gr
+import spaces
+import json
+import hashlib
+import logging
+from functools import lru_cache
+from transformers import pipeline
+
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger(__name__)
+
+_pipeline_cache = {"hash": None, "pipe": None, "config": None}
+
+def _hash_config(cfg: dict) -> str:
+    return hashlib.sha256(json.dumps(cfg, sort_keys=True).encode()).hexdigest()[:16]
+
+@lru_cache(maxsize=3)
+def _load_pipeline(hash_key: str, cfg_json: str):
+    cfg = json.loads(cfg_json)
+    logger.info(f"🔄 Loading pipeline: {cfg}")
+    pipe = pipeline(**cfg)
+    logger.info("✅ Loaded.")
+    return pipe
+
+def get_pipe(cfg: dict):
+    h = _hash_config(cfg)
+    if _pipeline_cache["hash"] == h and _pipeline_cache["pipe"] is not None:
+        return _pipeline_cache["pipe"], False
+    pipe = _load_pipeline(h, json.dumps(cfg, sort_keys=True))
+    _pipeline_cache.update({"hash": h, "pipe": pipe, "config": cfg})
+    return pipe, True
+
+@spaces.GPU(duration=40)
+def inference(pipeline_config, inputs, inference_kwargs=None):
+    if inference_kwargs is None:
+        inference_kwargs = {}
+    try:
+        pconf = json.loads(pipeline_config) if isinstance(pipeline_config, str) else pipeline_config
+    except Exception as e:
+        return json.dumps({"error": f"Bad pipeline_config: {e}"})
+    try:
+        ikw = json.loads(inference_kwargs) if isinstance(inference_kwargs, str) else inference_kwargs
+    except Exception as e:
+        return json.dumps({"error": f"Bad inference_kwargs: {e}"})
+
+    try:
+        pipe, reloaded = get_pipe(pconf)
+    except Exception as e:
+        return json.dumps({"error": f"Pipeline load failed: {e}"})
+
+    try:
+        raw = json.loads(inputs) if isinstance(inputs, str) and inputs.strip().startswith(("[", "{")) else inputs
+    except:
+        raw = inputs
+
+    try:
+        result = pipe(raw, **ikw)
+        return json.dumps({
+            "success": True,
+            "reloaded": reloaded,
+            "result": result,
+        }, default=str, indent=2)
+    except Exception as e:
+        return json.dumps({"error": f"Inference failed: {e}"})
+
+with gr.Blocks(title="Dynamic Transformers Pipeline API") as demo:
+    gr.Markdown("# 🚀 Dynamic Transformers Pipeline API\nZero-GPU. Pass any `transformers.pipeline` config via JSON.")
+    with gr.Row():
+        with gr.Column():
+            pcfg = gr.JSON(label="pipeline_config", value={"task": "text-generation", "model": "HuggingFaceTB/SmolLM2-135M-Instruct"})
+            inp = gr.Textbox(label="inputs", value="The future of AI is", lines=3)
+            ikw = gr.JSON(label="inference_kwargs", value={"max_new_tokens": 50})
+            btn = gr.Button("Run", variant="primary")
+        with gr.Column():
+            out = gr.JSON(label="output")
+    btn.click(inference, [pcfg, inp, ikw], out)
+
+    gr.Markdown("## API Example")
+    gr.Code("""from gradio_client import Client
+c = Client("your-username/your-space")
+print(c.predict(
+    pipeline_config={"task": "text-generation", "model": "HuggingFaceTB/SmolLM2-135M-Instruct"},
+    inputs="Hello",
+    inference_kwargs={"max_new_tokens": 20},
+    api_name="/inference"
+))""", language="python")
+
+demo.launch()

requirements.txt

@@ -0,0 +1,5 @@
+gradio>=4.0
+transformers
+torch
+accelerate
+spaces