Dec 1, 2023 · Dec 1, 2023 · Dec 1, 2023
diff --git a/pgml-extension/src/bindings/transformers/transformers.py b/pgml-extension/src/bindings/transformers/transformers.py
        self.next_tokens_are_prompt = True
        self.stop_signal = None
        self.text_queue = queue.Queue()
        self.token_cache = []
        self.text_index_cache = []

    def put(self,value):
    def put(self,values):
        if self.skip_prompt and self.next_tokens_are_prompt:
            self.next_tokens_are_prompt = False
            return
        # Can't batch this decode
        decoded_values = []
        for v in value:
            decoded_values.append(self.tokenizer.decode(v, **self.decode_kwargs))
        self.text_queue.put(decoded_values, self.timeout)
        output = []
        for i, v in enumerate(values):
            if len(self.token_cache) <= i:
                self.token_cache.append([])
                self.text_index_cache.append(0)
            token = v.tolist()  # Returns a list or number
            if type(token) == list:
                self.token_cache[i].extend(token)
            else:
                self.token_cache[i].append(token)
            text = self.tokenizer.decode(self.token_cache[i], **self.decode_kwargs)
            if text.endswith("\n"):
                output.append(text[self.text_index_cache[i] :])
                self.token_cache[i] = []
                self.text_index_cache[i] = 0
            else:
                printable_text = text[self.text_index_cache[i] : text.rfind(" ") + 1]
                self.text_index_cache[i] += len(printable_text)
                output.append(printable_text)
        if any(output):
            self.text_queue.put(output, self.timeout)

    def end(self):
        self.next_tokens_are_prompt = True
        output = []
        for i, tokens in enumerate(self.token_cache):
            text = self.tokenizer.decode(tokens, **self.decode_kwargs)
            output.append(text[self.text_index_cache[i] :])
        self.text_queue.put(output, self.timeout)
        self.text_queue.put(self.stop_signal, self.timeout)

    def __iter__(self):
        if value != self.stop_signal:
            return value


 class GGMLPipeline(object):
    def __init__(self, model_name, **task):
        import ctransformers
        generation_kwargs = None
        if self.task == "conversational":
            streamer = TextIteratorStreamer(
                self.tokenizer, skip_prompt=True, skip_special_tokens=True
                self.tokenizer,
                skip_prompt=True,
            )
            if "chat_template" in kwargs:
                input = self.tokenizer.apply_chat_template(
            input = self.tokenizer(input, return_tensors="pt").to(self.model.device)
            generation_kwargs = dict(input, streamer=streamer, **kwargs)
        else:
            streamer = TextIteratorStreamer(self.tokenizer, skip_special_tokens=True)
            streamer = TextIteratorStreamer(self.tokenizer)
            input = self.tokenizer(input, return_tensors="pt", padding=True).to(
                self.model.device
            )
diff --git a/pgml-sdks/pgml/build.rs b/pgml-sdks/pgml/build.rs

 Json = Any
 DateTime = int
 GeneralJsonIterator = Any
 GeneralJsonAsyncIterator = Any
 "#;

 const ADDITIONAL_DEFAULTS_FOR_#"9">

 export type Json = any;
 export type DateTime = Date;
 export type GeneralJsonIterator = any;
 export type GeneralJsonAsyncIterator = any;

 export function newCollection(name: string, database_url?: string): Collection;
 export function newModel(name?: string, source?: string, parameters?: Json): Model;
diff --git a/pgml-sdks/pgml/python/tests/test.py b/pgml-sdks/pgml/python/tests/test.py
        ],
        temperature=0.85,
    )
    import json
    assert len(results["choices"]) > 0


diff --git a/pgml-sdks/pgml/src/languages/javascript.rs b/pgml-sdks/pgml/src/languages/javascript.rs
                        .expect("Error converting rust Json to JavaScript Object");
                    let d = cx.boolean(false);
                    o.set(&mut cx, "value", v)
                        .expect("Error setting object value intransform_sream_iterate_next");
                        .expect("Error setting object value intransform_stream_iterate_next");
                    o.set(&mut cx, "done", d)
                        .expect("Error setting object value intransform_sream_iterate_next");
                        .expect("Error setting object value intransform_stream_iterate_next");
                } else {
                    let d = cx.boolean(true);
                    o.set(&mut cx, "done", d)
                        .expect("Error setting object value intransform_sream_iterate_next");
                        .expect("Error setting object value intransform_stream_iterate_next");
                }
                Ok(o)
            })
Original file line number	Diff line number	Diff line change
Expand Up		@@ -104,19 +104,42 @@ def __init__(self, tokenizer, skip_prompt=False, timeout=None, **decode_kwargs):
		self.next_tokens_are_prompt = True
		self.stop_signal = None
		self.text_queue = queue.Queue()
		self.token_cache = []
		self.text_index_cache = []

		def put(self,value):
		def put(self,values):
		if self.skip_prompt and self.next_tokens_are_prompt:
		self.next_tokens_are_prompt = False
		return
		# Can't batch this decode
		decoded_values = []
		for v in value:
		decoded_values.append(self.tokenizer.decode(v, **self.decode_kwargs))
		self.text_queue.put(decoded_values, self.timeout)
		output = []
		for i, v in enumerate(values):
		if len(self.token_cache) <= i:
		self.token_cache.append([])
		self.text_index_cache.append(0)
		token = v.tolist() # Returns a list or number
		if type(token) == list:
		self.token_cache[i].extend(token)
		else:
		self.token_cache[i].append(token)
		text = self.tokenizer.decode(self.token_cache[i], **self.decode_kwargs)
		if text.endswith("\n"):
		output.append(text[self.text_index_cache[i] :])
		self.token_cache[i] = []
		self.text_index_cache[i] = 0
		else:
		printable_text = text[self.text_index_cache[i] : text.rfind(" ") + 1]
		self.text_index_cache[i] += len(printable_text)
		output.append(printable_text)
		if any(output):
		self.text_queue.put(output, self.timeout)

		def end(self):
		self.next_tokens_are_prompt = True
		output = []
		for i, tokens in enumerate(self.token_cache):
		text = self.tokenizer.decode(tokens, **self.decode_kwargs)
		output.append(text[self.text_index_cache[i] :])
		self.text_queue.put(output, self.timeout)
		self.text_queue.put(self.stop_signal, self.timeout)

		def __iter__(self):
Expand All		@@ -127,6 +150,7 @@ def __next__(self):
		if value != self.stop_signal:
		return value


		class GGMLPipeline(object):
		def __init__(self, model_name, **task):
		import ctransformers
Expand DownExpand Up		@@ -245,7 +269,8 @@ def stream(self, input, **kwargs):
		generation_kwargs = None
		if self.task == "conversational":
		streamer = TextIteratorStreamer(
		self.tokenizer, skip_prompt=True, skip_special_tokens=True
		self.tokenizer,
		skip_prompt=True,
		)
		if "chat_template" in kwargs:
		input = self.tokenizer.apply_chat_template(
Expand All		@@ -261,7 +286,7 @@ def stream(self, input, **kwargs):
		input = self.tokenizer(input, return_tensors="pt").to(self.model.device)
		generation_kwargs = dict(input, streamer=streamer, **kwargs)
		else:
		streamer = TextIteratorStreamer(self.tokenizer, skip_special_tokens=True)
		streamer = TextIteratorStreamer(self.tokenizer)
		input = self.tokenizer(input, return_tensors="pt", padding=True).to(
		self.model.device
		)
Expand Down
Original file line number	Diff line change
Expand Up	@@ -8,6 +8,8 @@ async def migrate() -> None

	Json = Any
	DateTime = int
	GeneralJsonIterator = Any
	GeneralJsonAsyncIterator = Any
	"#;

	const ADDITIONAL_DEFAULTS_FOR_#"9">	Expand All	@@ -16,6 +18,8 @@ export function migrate(): Promise<void>;

	export type Json = any;
	export type DateTime = Date;
	export type GeneralJsonIterator = any;
	export type GeneralJsonAsyncIterator = any;

	export function newCollection(name: string, database_url?: string): Collection;
	export function newModel(name?: string, source?: string, parameters?: Json): Model;
Expand Down
Original file line number	Diff line number	Diff line change
Expand Up		@@ -361,7 +361,6 @@ async def test_open_source_ai_create_async():
		],
		temperature=0.85,
		)
		import json
		assert len(results["choices"]) > 0


Expand Down
Original file line number	Diff line number	Diff line change
Expand Up		@@ -101,13 +101,13 @@ fn transform_stream_iterate_next(mut cx: FunctionContext) -> JsResult<JsPromise>
		.expect("Error converting rust Json to JavaScript Object");
		let d = cx.boolean(false);
		o.set(&mut cx, "value", v)
		.expect("Error setting object value intransform_sream_iterate_next");
		.expect("Error setting object value intransform_stream_iterate_next");
		o.set(&mut cx, "done", d)
		.expect("Error setting object value intransform_sream_iterate_next");
		.expect("Error setting object value intransform_stream_iterate_next");
		} else {
		let d = cx.boolean(true);
		o.set(&mut cx, "done", d)
		.expect("Error setting object value intransform_sream_iterate_next");
		.expect("Error setting object value intransform_stream_iterate_next");
		}
		Ok(o)
		})
Expand Down