feat(api): add txt2txt endpoint

api/onnx_web/serve.py

@@ -64,6 +64,7 @@
     UpscaleParams,
 )
 from .server import DevicePoolExecutor, ServerContext, apply_patches
+from .transformers import run_txt2txt_pipeline
 from .utils import (
     base_join,
     get_and_clamp_float,
@@ -809,6 +810,26 @@ def blend():
     return jsonify(json_params(output, params, size, upscale=upscale))
 
 
+@app.route("/api/txt2txt", methods=["POST"])
+def txt2txt():
+    device, params, size = pipeline_from_request()
+
+    output = make_output_name(context, "upscale", params, size)
+    logger.info("upscale job queued for: %s", output)
+
+    executor.submit(
+        output,
+        run_txt2txt_pipeline,
+        context,
+        params,
+        size,
+        output,
+        needs_device=device,
+    )
+
+    return jsonify(json_params(output, params, size))
+
+
 @app.route("/api/cancel", methods=["PUT"])
 def cancel():
     output_file = request.args.get("output", None)

api/onnx_web/transformers.py

@@ -0,0 +1,40 @@
+from logging import getLogger
+
+from .params import ImageParams, Size
+from .server import JobContext, ServerContext
+
+logger = getLogger(__name__)
+
+
+def run_txt2txt_pipeline(
+    job: JobContext,
+    _server: ServerContext,
+    params: ImageParams,
+    _size: Size,
+    output: str,
+) -> None:
+    from transformers import AutoTokenizer, GPTJForCausalLM
+
+    # tested with "EleutherAI/gpt-j-6B"
+    model = "EleutherAI/gpt-j-6B"
+    tokens = 1024
+
+    device = job.get_device()
+
+    model = GPTJForCausalLM.from_pretrained(model).to(device.torch_device())
+    tokenizer = AutoTokenizer.from_pretrained(model)
+
+    input_ids = tokenizer.encode(params.prompt, return_tensors="pt").to(
+        device.torch_device()
+    )
+    output = model.generate(
+        input_ids,
+        do_sample=True,
+        max_length=tokens,
+        temperature=0.8,
+    )
+    result = tokenizer.decode(output[0], skip_special_tokens=True)
+
+    print("Server says: %s" % result)
+
+    logger.info("finished txt2txt job: %s", output)