Upload HfMoondream
Browse files- config.json +1 -1
- generation_config.json +1 -1
- model.safetensors +1 -1
- moondream.py +101 -38
config.json
CHANGED
@@ -9,5 +9,5 @@
|
|
9 |
"config": {},
|
10 |
"model_type": "moondream1",
|
11 |
"torch_dtype": "float16",
|
12 |
-
"transformers_version": "4.
|
13 |
}
|
|
|
9 |
"config": {},
|
10 |
"model_type": "moondream1",
|
11 |
"torch_dtype": "float16",
|
12 |
+
"transformers_version": "4.44.0"
|
13 |
}
|
generation_config.json
CHANGED
@@ -1,4 +1,4 @@
|
|
1 |
{
|
2 |
"_from_model_config": true,
|
3 |
-
"transformers_version": "4.
|
4 |
}
|
|
|
1 |
{
|
2 |
"_from_model_config": true,
|
3 |
+
"transformers_version": "4.44.0"
|
4 |
}
|
model.safetensors
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 3854538376
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:96dce588e4a319fde7af3c70fbf27e726f4850e22522d0fdc4b165d5e6003ad5
|
3 |
size 3854538376
|
moondream.py
CHANGED
@@ -15,13 +15,26 @@ from .region import decode_coordinate, encode_coordinate, decode_size, encode_si
|
|
15 |
from .utils import remove_outlier_points
|
16 |
|
17 |
|
18 |
-
|
19 |
-
"
|
20 |
-
{
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
total=False,
|
22 |
)
|
23 |
|
24 |
DEFAULT_MAX_TOKENS = 768
|
|
|
|
|
|
|
25 |
|
26 |
|
27 |
@dataclass(frozen=True)
|
@@ -144,7 +157,7 @@ class MoondreamModel(nn.Module):
|
|
144 |
def _decode_one_tok(
|
145 |
self, x: torch.Tensor, attn_mask: torch.Tensor, pos_ids: torch.Tensor
|
146 |
):
|
147 |
-
hidden = text_decoder(x
|
148 |
logits = lm_head(hidden, self.text)
|
149 |
return logits, hidden
|
150 |
|
@@ -209,7 +222,19 @@ class MoondreamModel(nn.Module):
|
|
209 |
],
|
210 |
)
|
211 |
|
212 |
-
def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
213 |
with torch.inference_mode():
|
214 |
prompt_emb = text_encoder(prompt_tokens, self.text)
|
215 |
torch._dynamo.mark_dynamic(prompt_emb, 1)
|
@@ -217,7 +242,14 @@ class MoondreamModel(nn.Module):
|
|
217 |
pos_ids = torch.arange(pos, pos + prompt_emb.size(1), dtype=torch.long)
|
218 |
hidden = self._prefill(prompt_emb, mask, pos_ids)
|
219 |
logits = lm_head(hidden, self.text)
|
220 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
221 |
pos = pos + prompt_emb.size(1)
|
222 |
return logits, hidden, next_token, pos
|
223 |
|
@@ -225,9 +257,23 @@ class MoondreamModel(nn.Module):
|
|
225 |
self,
|
226 |
prompt_tokens: torch.Tensor,
|
227 |
pos: int,
|
228 |
-
|
229 |
):
|
230 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
231 |
|
232 |
def generator(next_token, pos):
|
233 |
mask = torch.zeros(1, 1, 2048, device=self.device, dtype=torch.bool)
|
@@ -275,7 +321,14 @@ class MoondreamModel(nn.Module):
|
|
275 |
mask[:, :, pos], pos_ids[0] = 1, pos
|
276 |
logits, _ = self._decode_one_tok(next_emb, mask, pos_ids)
|
277 |
pos += 1
|
278 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
279 |
generated_tokens += 1
|
280 |
|
281 |
# Flush any remaining text in the cache
|
@@ -292,7 +345,7 @@ class MoondreamModel(nn.Module):
|
|
292 |
image: Union[Image.Image, EncodedImage],
|
293 |
question: str,
|
294 |
stream: bool = False,
|
295 |
-
settings: Optional[
|
296 |
):
|
297 |
if self.config.tokenizer.templates["query"] is None:
|
298 |
raise NotImplementedError("Model does not support querying.")
|
@@ -309,12 +362,8 @@ class MoondreamModel(nn.Module):
|
|
309 |
device=self.device,
|
310 |
)
|
311 |
|
312 |
-
max_tokens = DEFAULT_MAX_TOKENS
|
313 |
-
if settings:
|
314 |
-
max_tokens = settings.get("max_tokens", DEFAULT_MAX_TOKENS)
|
315 |
-
|
316 |
def generator():
|
317 |
-
for token in self._generate_text(prompt_tokens, image.pos,
|
318 |
yield token
|
319 |
|
320 |
if stream:
|
@@ -332,7 +381,7 @@ class MoondreamModel(nn.Module):
|
|
332 |
image: Union[Image.Image, EncodedImage],
|
333 |
length: Literal["normal", "short", "long"] = "normal",
|
334 |
stream: bool = False,
|
335 |
-
settings: Optional[
|
336 |
):
|
337 |
if self.config.tokenizer.templates["caption"] is None:
|
338 |
raise NotImplementedError("Model does not support captioning.")
|
@@ -346,12 +395,8 @@ class MoondreamModel(nn.Module):
|
|
346 |
[self.config.tokenizer.templates["caption"][length]], device=self.device
|
347 |
)
|
348 |
|
349 |
-
max_tokens = DEFAULT_MAX_TOKENS
|
350 |
-
if settings:
|
351 |
-
max_tokens = settings.get("max_tokens", DEFAULT_MAX_TOKENS)
|
352 |
-
|
353 |
def generator():
|
354 |
-
for token in self._generate_text(prompt_tokens, image.pos,
|
355 |
yield token
|
356 |
|
357 |
if stream:
|
@@ -365,7 +410,7 @@ class MoondreamModel(nn.Module):
|
|
365 |
next_token: torch.Tensor,
|
366 |
pos: int,
|
367 |
include_size: bool = True,
|
368 |
-
|
369 |
):
|
370 |
out = []
|
371 |
mask = torch.zeros(1, 1, 2048, device=self.device, dtype=torch.bool)
|
@@ -375,13 +420,13 @@ class MoondreamModel(nn.Module):
|
|
375 |
with torch.inference_mode():
|
376 |
while (
|
377 |
next_token.item() != self.config.tokenizer.eos_id
|
378 |
-
and len(out) <
|
379 |
):
|
380 |
x_logits = decode_coordinate(hidden, self.region)
|
381 |
x_center = torch.argmax(x_logits, dim=-1) / x_logits.size(-1)
|
382 |
next_emb = encode_coordinate(
|
383 |
x_center.to(dtype=x_logits.dtype), self.region
|
384 |
-
)
|
385 |
|
386 |
# Decode y-coordinate
|
387 |
mask[:, :, pos], pos_ids[0] = 1, pos
|
@@ -391,7 +436,7 @@ class MoondreamModel(nn.Module):
|
|
391 |
y_center = torch.argmax(y_logits, dim=-1) / y_logits.size(-1)
|
392 |
next_emb = encode_coordinate(
|
393 |
y_center.to(dtype=y_logits.dtype), self.region
|
394 |
-
)
|
395 |
|
396 |
# Decode size
|
397 |
if include_size:
|
@@ -409,12 +454,16 @@ class MoondreamModel(nn.Module):
|
|
409 |
w = torch.pow(2.0, (w_bin.float() / 1023.0) * 10.0 - 10.0)
|
410 |
h = torch.pow(2.0, (h_bin.float() / 1023.0) * 10.0 - 10.0)
|
411 |
|
412 |
-
next_emb =
|
413 |
-
|
414 |
-
|
415 |
-
|
416 |
-
|
417 |
-
|
|
|
|
|
|
|
|
|
418 |
|
419 |
# Add object
|
420 |
out.append(
|
@@ -440,7 +489,7 @@ class MoondreamModel(nn.Module):
|
|
440 |
self,
|
441 |
image: Union[Image.Image, EncodedImage],
|
442 |
object: str,
|
443 |
-
settings: Optional[
|
444 |
):
|
445 |
if self.config.tokenizer.templates["detect"] is None:
|
446 |
raise NotImplementedError("Model does not support object detection.")
|
@@ -457,11 +506,18 @@ class MoondreamModel(nn.Module):
|
|
457 |
device=self.device,
|
458 |
)
|
459 |
|
460 |
-
_, hidden, next_token, pos = self._prefill_prompt(
|
|
|
|
|
461 |
hidden = hidden[:, -1:, :]
|
462 |
|
|
|
|
|
|
|
|
|
|
|
463 |
objects = self._generate_points(
|
464 |
-
hidden, next_token, pos, include_size=True,
|
465 |
)
|
466 |
|
467 |
return {"objects": objects}
|
@@ -470,7 +526,7 @@ class MoondreamModel(nn.Module):
|
|
470 |
self,
|
471 |
image: Union[Image.Image, EncodedImage],
|
472 |
object: str,
|
473 |
-
settings: Optional[
|
474 |
):
|
475 |
if self.config.tokenizer.templates["point"] is None:
|
476 |
raise NotImplementedError("Model does not support pointing.")
|
@@ -487,11 +543,18 @@ class MoondreamModel(nn.Module):
|
|
487 |
device=self.device,
|
488 |
)
|
489 |
|
490 |
-
_, hidden, next_token, pos = self._prefill_prompt(
|
|
|
|
|
491 |
hidden = hidden[:, -1:, :]
|
492 |
|
|
|
|
|
|
|
|
|
|
|
493 |
objects = self._generate_points(
|
494 |
-
hidden, next_token, pos, include_size=False,
|
495 |
)
|
496 |
|
497 |
return {"points": objects}
|
@@ -545,7 +608,7 @@ class MoondreamModel(nn.Module):
|
|
545 |
return None
|
546 |
|
547 |
gaze = self._generate_points(
|
548 |
-
hidden, next_token, pos, include_size=False,
|
549 |
)
|
550 |
return gaze[0]
|
551 |
|
|
|
15 |
from .utils import remove_outlier_points
|
16 |
|
17 |
|
18 |
+
TextSamplingSettings = TypedDict(
|
19 |
+
"TextSamplingSettings",
|
20 |
+
{
|
21 |
+
"max_tokens": int,
|
22 |
+
"temperature": float,
|
23 |
+
"top_p": float,
|
24 |
+
},
|
25 |
+
total=False,
|
26 |
+
)
|
27 |
+
|
28 |
+
ObjectSamplingSettings = TypedDict(
|
29 |
+
"ObjectSamplingSettings",
|
30 |
+
{"max_objects": int},
|
31 |
total=False,
|
32 |
)
|
33 |
|
34 |
DEFAULT_MAX_TOKENS = 768
|
35 |
+
DEFAULT_TEMPERATURE = 0.5
|
36 |
+
DEFAULT_TOP_P = 0.3
|
37 |
+
DEFAULT_MAX_OBJECTS = 50
|
38 |
|
39 |
|
40 |
@dataclass(frozen=True)
|
|
|
157 |
def _decode_one_tok(
|
158 |
self, x: torch.Tensor, attn_mask: torch.Tensor, pos_ids: torch.Tensor
|
159 |
):
|
160 |
+
hidden = text_decoder(x, self.text, attn_mask, pos_ids, self.config.text)
|
161 |
logits = lm_head(hidden, self.text)
|
162 |
return logits, hidden
|
163 |
|
|
|
222 |
],
|
223 |
)
|
224 |
|
225 |
+
def _apply_top_p(self, probs: torch.Tensor, top_p: float):
|
226 |
+
probs_sort, probs_idx = torch.sort(probs, dim=-1, descending=True)
|
227 |
+
probs_sum = torch.cumsum(probs_sort, dim=-1)
|
228 |
+
mask = probs_sum - probs_sort > top_p
|
229 |
+
probs_sort[mask] = 0.0
|
230 |
+
probs_sort.div_(probs_sort.sum(dim=-1, keepdim=True))
|
231 |
+
next_probs = torch.zeros_like(probs)
|
232 |
+
next_probs.scatter_(dim=-1, index=probs_idx, src=probs_sort)
|
233 |
+
return next_probs
|
234 |
+
|
235 |
+
def _prefill_prompt(
|
236 |
+
self, prompt_tokens: torch.Tensor, pos: int, temperature: float, top_p: float
|
237 |
+
):
|
238 |
with torch.inference_mode():
|
239 |
prompt_emb = text_encoder(prompt_tokens, self.text)
|
240 |
torch._dynamo.mark_dynamic(prompt_emb, 1)
|
|
|
242 |
pos_ids = torch.arange(pos, pos + prompt_emb.size(1), dtype=torch.long)
|
243 |
hidden = self._prefill(prompt_emb, mask, pos_ids)
|
244 |
logits = lm_head(hidden, self.text)
|
245 |
+
|
246 |
+
if temperature == 0:
|
247 |
+
next_token = torch.argmax(logits, dim=-1).unsqueeze(1)
|
248 |
+
else:
|
249 |
+
probs = torch.softmax(logits / temperature, dim=-1)
|
250 |
+
probs = self._apply_top_p(probs, top_p)
|
251 |
+
next_token = torch.multinomial(probs, num_samples=1)
|
252 |
+
|
253 |
pos = pos + prompt_emb.size(1)
|
254 |
return logits, hidden, next_token, pos
|
255 |
|
|
|
257 |
self,
|
258 |
prompt_tokens: torch.Tensor,
|
259 |
pos: int,
|
260 |
+
settings: Optional[TextSamplingSettings] = None,
|
261 |
):
|
262 |
+
max_tokens = (
|
263 |
+
settings.get("max_tokens", DEFAULT_MAX_TOKENS)
|
264 |
+
if settings
|
265 |
+
else DEFAULT_MAX_TOKENS
|
266 |
+
)
|
267 |
+
temperature = (
|
268 |
+
settings.get("temperature", DEFAULT_TEMPERATURE)
|
269 |
+
if settings
|
270 |
+
else DEFAULT_TEMPERATURE
|
271 |
+
)
|
272 |
+
top_p = settings.get("top_p", DEFAULT_TOP_P) if settings else DEFAULT_TOP_P
|
273 |
+
|
274 |
+
_, _, next_token, pos = self._prefill_prompt(
|
275 |
+
prompt_tokens, pos, temperature, top_p
|
276 |
+
)
|
277 |
|
278 |
def generator(next_token, pos):
|
279 |
mask = torch.zeros(1, 1, 2048, device=self.device, dtype=torch.bool)
|
|
|
321 |
mask[:, :, pos], pos_ids[0] = 1, pos
|
322 |
logits, _ = self._decode_one_tok(next_emb, mask, pos_ids)
|
323 |
pos += 1
|
324 |
+
|
325 |
+
if temperature == 0:
|
326 |
+
next_token = torch.argmax(logits, dim=-1).unsqueeze(1) # (1, 1)
|
327 |
+
else:
|
328 |
+
probs = torch.softmax(logits / temperature, dim=-1) # (1, V)
|
329 |
+
probs = self._apply_top_p(probs, top_p)
|
330 |
+
next_token = torch.multinomial(probs, num_samples=1) # (1, 1)
|
331 |
+
|
332 |
generated_tokens += 1
|
333 |
|
334 |
# Flush any remaining text in the cache
|
|
|
345 |
image: Union[Image.Image, EncodedImage],
|
346 |
question: str,
|
347 |
stream: bool = False,
|
348 |
+
settings: Optional[TextSamplingSettings] = None,
|
349 |
):
|
350 |
if self.config.tokenizer.templates["query"] is None:
|
351 |
raise NotImplementedError("Model does not support querying.")
|
|
|
362 |
device=self.device,
|
363 |
)
|
364 |
|
|
|
|
|
|
|
|
|
365 |
def generator():
|
366 |
+
for token in self._generate_text(prompt_tokens, image.pos, settings):
|
367 |
yield token
|
368 |
|
369 |
if stream:
|
|
|
381 |
image: Union[Image.Image, EncodedImage],
|
382 |
length: Literal["normal", "short", "long"] = "normal",
|
383 |
stream: bool = False,
|
384 |
+
settings: Optional[TextSamplingSettings] = None,
|
385 |
):
|
386 |
if self.config.tokenizer.templates["caption"] is None:
|
387 |
raise NotImplementedError("Model does not support captioning.")
|
|
|
395 |
[self.config.tokenizer.templates["caption"][length]], device=self.device
|
396 |
)
|
397 |
|
|
|
|
|
|
|
|
|
398 |
def generator():
|
399 |
+
for token in self._generate_text(prompt_tokens, image.pos, settings):
|
400 |
yield token
|
401 |
|
402 |
if stream:
|
|
|
410 |
next_token: torch.Tensor,
|
411 |
pos: int,
|
412 |
include_size: bool = True,
|
413 |
+
max_objects: int = DEFAULT_MAX_OBJECTS,
|
414 |
):
|
415 |
out = []
|
416 |
mask = torch.zeros(1, 1, 2048, device=self.device, dtype=torch.bool)
|
|
|
420 |
with torch.inference_mode():
|
421 |
while (
|
422 |
next_token.item() != self.config.tokenizer.eos_id
|
423 |
+
and len(out) < max_objects
|
424 |
):
|
425 |
x_logits = decode_coordinate(hidden, self.region)
|
426 |
x_center = torch.argmax(x_logits, dim=-1) / x_logits.size(-1)
|
427 |
next_emb = encode_coordinate(
|
428 |
x_center.to(dtype=x_logits.dtype), self.region
|
429 |
+
).unsqueeze(0)
|
430 |
|
431 |
# Decode y-coordinate
|
432 |
mask[:, :, pos], pos_ids[0] = 1, pos
|
|
|
436 |
y_center = torch.argmax(y_logits, dim=-1) / y_logits.size(-1)
|
437 |
next_emb = encode_coordinate(
|
438 |
y_center.to(dtype=y_logits.dtype), self.region
|
439 |
+
).unsqueeze(0)
|
440 |
|
441 |
# Decode size
|
442 |
if include_size:
|
|
|
454 |
w = torch.pow(2.0, (w_bin.float() / 1023.0) * 10.0 - 10.0)
|
455 |
h = torch.pow(2.0, (h_bin.float() / 1023.0) * 10.0 - 10.0)
|
456 |
|
457 |
+
next_emb = (
|
458 |
+
encode_size(
|
459 |
+
torch.tensor(
|
460 |
+
[w, h], device=self.device, dtype=size_logits.dtype
|
461 |
+
),
|
462 |
+
self.region,
|
463 |
+
)
|
464 |
+
.unsqueeze(0)
|
465 |
+
.unsqueeze(0)
|
466 |
+
)
|
467 |
|
468 |
# Add object
|
469 |
out.append(
|
|
|
489 |
self,
|
490 |
image: Union[Image.Image, EncodedImage],
|
491 |
object: str,
|
492 |
+
settings: Optional[ObjectSamplingSettings] = None,
|
493 |
):
|
494 |
if self.config.tokenizer.templates["detect"] is None:
|
495 |
raise NotImplementedError("Model does not support object detection.")
|
|
|
506 |
device=self.device,
|
507 |
)
|
508 |
|
509 |
+
_, hidden, next_token, pos = self._prefill_prompt(
|
510 |
+
prompt_tokens, image.pos, temperature=0, top_p=0
|
511 |
+
)
|
512 |
hidden = hidden[:, -1:, :]
|
513 |
|
514 |
+
max_objects = (
|
515 |
+
settings.get("max_objects", DEFAULT_MAX_OBJECTS)
|
516 |
+
if settings
|
517 |
+
else DEFAULT_MAX_OBJECTS
|
518 |
+
)
|
519 |
objects = self._generate_points(
|
520 |
+
hidden, next_token, pos, include_size=True, max_objects=max_objects
|
521 |
)
|
522 |
|
523 |
return {"objects": objects}
|
|
|
526 |
self,
|
527 |
image: Union[Image.Image, EncodedImage],
|
528 |
object: str,
|
529 |
+
settings: Optional[ObjectSamplingSettings] = None,
|
530 |
):
|
531 |
if self.config.tokenizer.templates["point"] is None:
|
532 |
raise NotImplementedError("Model does not support pointing.")
|
|
|
543 |
device=self.device,
|
544 |
)
|
545 |
|
546 |
+
_, hidden, next_token, pos = self._prefill_prompt(
|
547 |
+
prompt_tokens, image.pos, temperature=0, top_p=0
|
548 |
+
)
|
549 |
hidden = hidden[:, -1:, :]
|
550 |
|
551 |
+
max_objects = (
|
552 |
+
settings.get("max_objects", DEFAULT_MAX_OBJECTS)
|
553 |
+
if settings
|
554 |
+
else DEFAULT_MAX_OBJECTS
|
555 |
+
)
|
556 |
objects = self._generate_points(
|
557 |
+
hidden, next_token, pos, include_size=False, max_objects=max_objects
|
558 |
)
|
559 |
|
560 |
return {"points": objects}
|
|
|
608 |
return None
|
609 |
|
610 |
gaze = self._generate_points(
|
611 |
+
hidden, next_token, pos, include_size=False, max_objects=1
|
612 |
)
|
613 |
return gaze[0]
|
614 |
|