Upload HfMoondream
Browse files- config.json +2 -2
- config.py +16 -8
- hf_moondream.py +46 -5
- image_crops.py +36 -13
- layers.py +109 -6
- lora.py +82 -0
- model.safetensors +2 -2
- moondream.py +330 -61
- region.py +50 -3
- text.py +34 -18
- vision.py +3 -3
config.json
CHANGED
@@ -8,6 +8,6 @@
|
|
8 |
},
|
9 |
"config": {},
|
10 |
"model_type": "moondream1",
|
11 |
-
"torch_dtype": "
|
12 |
-
"transformers_version": "4.
|
13 |
}
|
|
|
8 |
},
|
9 |
"config": {},
|
10 |
"model_type": "moondream1",
|
11 |
+
"torch_dtype": "bfloat16",
|
12 |
+
"transformers_version": "4.52.4"
|
13 |
}
|
config.py
CHANGED
@@ -12,6 +12,7 @@ class TextConfig:
|
|
12 |
n_heads: int = 32
|
13 |
n_kv_heads: int = 32
|
14 |
prefix_attn: int = 730
|
|
|
15 |
|
16 |
|
17 |
@dataclass(frozen=True)
|
@@ -37,22 +38,29 @@ class RegionConfig:
|
|
37 |
size_feat_dim: int = 512
|
38 |
size_out_dim: int = 2048
|
39 |
inner_dim: int = 8192
|
|
|
40 |
|
41 |
|
42 |
@dataclass(frozen=True)
|
43 |
class TokenizerConfig:
|
44 |
-
bos_id: int =
|
45 |
-
eos_id: int =
|
|
|
|
|
|
|
|
|
|
|
|
|
46 |
templates: Dict[str, Optional[Dict[str, List[int]]]] = field(
|
47 |
default_factory=lambda: {
|
48 |
"caption": {
|
49 |
-
"short": [
|
50 |
-
"normal": [
|
51 |
-
"long": [
|
52 |
},
|
53 |
-
"query": {"prefix": [
|
54 |
-
"detect": {"prefix": [
|
55 |
-
"point": {"prefix": [
|
56 |
}
|
57 |
)
|
58 |
|
|
|
12 |
n_heads: int = 32
|
13 |
n_kv_heads: int = 32
|
14 |
prefix_attn: int = 730
|
15 |
+
group_size: Optional[int] = None
|
16 |
|
17 |
|
18 |
@dataclass(frozen=True)
|
|
|
38 |
size_feat_dim: int = 512
|
39 |
size_out_dim: int = 2048
|
40 |
inner_dim: int = 8192
|
41 |
+
group_size: Optional[int] = None
|
42 |
|
43 |
|
44 |
@dataclass(frozen=True)
|
45 |
class TokenizerConfig:
|
46 |
+
bos_id: int = 0
|
47 |
+
eos_id: int = 0
|
48 |
+
answer_id: int = 3
|
49 |
+
thinking_id: int = 4
|
50 |
+
coord_id: int = 5
|
51 |
+
size_id: int = 6
|
52 |
+
start_ground_points_id: int = 7
|
53 |
+
end_ground_id: int = 9
|
54 |
templates: Dict[str, Optional[Dict[str, List[int]]]] = field(
|
55 |
default_factory=lambda: {
|
56 |
"caption": {
|
57 |
+
"short": [1, 32708, 2, 12492, 3],
|
58 |
+
"normal": [1, 32708, 2, 6382, 3],
|
59 |
+
"long": [1, 32708, 2, 4059, 3],
|
60 |
},
|
61 |
+
"query": {"prefix": [1, 15381, 2], "suffix": [3]},
|
62 |
+
"detect": {"prefix": [1, 7235, 476, 2], "suffix": [3]},
|
63 |
+
"point": {"prefix": [1, 2581, 2], "suffix": [3]},
|
64 |
}
|
65 |
)
|
66 |
|
hf_moondream.py
CHANGED
@@ -1,4 +1,8 @@
|
|
|
|
|
|
|
|
1 |
from transformers import PreTrainedModel, PretrainedConfig
|
|
|
2 |
|
3 |
from .config import MoondreamConfig
|
4 |
from .moondream import MoondreamModel
|
@@ -123,7 +127,7 @@ class HfMoondream(PreTrainedModel):
|
|
123 |
)
|
124 |
|
125 |
def generator():
|
126 |
-
for token in self.model.
|
127 |
prompt_tokens,
|
128 |
image_embeds.kv_cache,
|
129 |
image_embeds.pos,
|
@@ -135,8 +139,45 @@ class HfMoondream(PreTrainedModel):
|
|
135 |
|
136 |
return [answer]
|
137 |
|
138 |
-
def get_input_embeddings(self):
|
139 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
140 |
|
141 |
-
|
142 |
-
self._unsupported_exception()
|
|
|
1 |
+
import torch
|
2 |
+
import torch.nn as nn
|
3 |
+
|
4 |
from transformers import PreTrainedModel, PretrainedConfig
|
5 |
+
from typing import Union
|
6 |
|
7 |
from .config import MoondreamConfig
|
8 |
from .moondream import MoondreamModel
|
|
|
127 |
)
|
128 |
|
129 |
def generator():
|
130 |
+
for token in self.model._generate_answer(
|
131 |
prompt_tokens,
|
132 |
image_embeds.kv_cache,
|
133 |
image_embeds.pos,
|
|
|
139 |
|
140 |
return [answer]
|
141 |
|
142 |
+
def get_input_embeddings(self) -> nn.Embedding:
|
143 |
+
"""
|
144 |
+
Lazily wrap the raw parameter `self.model.text.wte` in a real
|
145 |
+
`nn.Embedding` layer so that HF mix-ins recognise it. The wrapper
|
146 |
+
**shares** the weight tensor—no copy is made.
|
147 |
+
"""
|
148 |
+
if not hasattr(self, "_input_embeddings"):
|
149 |
+
self._input_embeddings = nn.Embedding.from_pretrained(
|
150 |
+
self.model.text.wte, # tensor created in text.py
|
151 |
+
freeze=True, # set to False if you need it trainable
|
152 |
+
)
|
153 |
+
return self._input_embeddings
|
154 |
+
|
155 |
+
def set_input_embeddings(self, value: Union[nn.Embedding, nn.Module]) -> None:
|
156 |
+
"""
|
157 |
+
Lets HF functions (e.g. `resize_token_embeddings`) replace or resize the
|
158 |
+
embeddings and keeps everything tied to `self.model.text.wte`.
|
159 |
+
"""
|
160 |
+
# 1. point the low-level parameter to the new weight matrix
|
161 |
+
self.model.text.wte = value.weight
|
162 |
+
# 2. keep a reference for get_input_embeddings()
|
163 |
+
self._input_embeddings = value
|
164 |
+
|
165 |
+
def input_embeds(
|
166 |
+
self,
|
167 |
+
input_ids: Union[torch.LongTensor, list, tuple],
|
168 |
+
*,
|
169 |
+
device: torch.device | None = None
|
170 |
+
) -> torch.FloatTensor:
|
171 |
+
"""
|
172 |
+
Back-compat wrapper that turns token IDs into embeddings.
|
173 |
+
|
174 |
+
Example:
|
175 |
+
ids = torch.tensor([[1, 2, 3]])
|
176 |
+
embeds = model.input_embeds(ids) # (1, 3, hidden_dim)
|
177 |
+
"""
|
178 |
+
if not torch.is_tensor(input_ids):
|
179 |
+
input_ids = torch.as_tensor(input_ids)
|
180 |
+
if device is not None:
|
181 |
+
input_ids = input_ids.to(device)
|
182 |
|
183 |
+
return self.get_input_embeddings()(input_ids)
|
|
image_crops.py
CHANGED
@@ -1,10 +1,18 @@
|
|
1 |
import math
|
2 |
import numpy as np
|
3 |
import torch
|
4 |
-
import pyvips
|
5 |
|
6 |
from typing import TypedDict
|
7 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
|
9 |
def select_tiling(
|
10 |
height: int, width: int, crop_size: int, max_crops: int
|
@@ -113,18 +121,33 @@ def overlap_crop_image(
|
|
113 |
tiling[1] * crop_window_size + total_margin_pixels,
|
114 |
)
|
115 |
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
128 |
|
129 |
for i in range(tiling[0]):
|
130 |
for j in range(tiling[1]):
|
|
|
1 |
import math
|
2 |
import numpy as np
|
3 |
import torch
|
|
|
4 |
|
5 |
from typing import TypedDict
|
6 |
|
7 |
+
try:
|
8 |
+
import pyvips
|
9 |
+
|
10 |
+
HAS_VIPS = True
|
11 |
+
except:
|
12 |
+
from PIL import Image
|
13 |
+
|
14 |
+
HAS_VIPS = False
|
15 |
+
|
16 |
|
17 |
def select_tiling(
|
18 |
height: int, width: int, crop_size: int, max_crops: int
|
|
|
121 |
tiling[1] * crop_window_size + total_margin_pixels,
|
122 |
)
|
123 |
|
124 |
+
if HAS_VIPS:
|
125 |
+
# Convert to vips for resizing
|
126 |
+
vips_image = pyvips.Image.new_from_array(image)
|
127 |
+
scale_x = target_size[1] / image.shape[1]
|
128 |
+
scale_y = target_size[0] / image.shape[0]
|
129 |
+
resized = vips_image.resize(scale_x, vscale=scale_y)
|
130 |
+
image = resized.numpy()
|
131 |
+
|
132 |
+
# Create global crop
|
133 |
+
scale_x = base_size[1] / vips_image.width
|
134 |
+
scale_y = base_size[0] / vips_image.height
|
135 |
+
global_vips = vips_image.resize(scale_x, vscale=scale_y)
|
136 |
+
crops[0] = global_vips.numpy()
|
137 |
+
else:
|
138 |
+
# Fallback to PIL
|
139 |
+
pil_img = Image.fromarray(image)
|
140 |
+
resized = pil_img.resize(
|
141 |
+
(int(target_size[1]), int(target_size[0])),
|
142 |
+
resample=Image.Resampling.LANCZOS,
|
143 |
+
)
|
144 |
+
image = np.asarray(resized)
|
145 |
+
|
146 |
+
# Create global crop
|
147 |
+
global_pil = pil_img.resize(
|
148 |
+
(int(base_size[1]), int(base_size[0])), resample=Image.Resampling.LANCZOS
|
149 |
+
)
|
150 |
+
crops[0] = np.asarray(global_pil)
|
151 |
|
152 |
for i in range(tiling[0]):
|
153 |
for j in range(tiling[1]):
|
layers.py
CHANGED
@@ -1,8 +1,24 @@
|
|
|
|
|
|
|
|
|
|
1 |
from dataclasses import dataclass
|
2 |
-
from typing import Literal
|
3 |
|
4 |
-
|
5 |
-
from
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
|
7 |
|
8 |
def gelu_approx(x):
|
@@ -19,6 +35,80 @@ def linear(x: torch.Tensor, w: LinearWeights) -> torch.Tensor:
|
|
19 |
return F.linear(x, w.weight, w.bias)
|
20 |
|
21 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
22 |
@dataclass
|
23 |
class LayerNormWeights:
|
24 |
weight: torch.Tensor
|
@@ -36,10 +126,23 @@ class MLPWeights:
|
|
36 |
act: Literal["gelu_approx"] = "gelu_approx"
|
37 |
|
38 |
|
39 |
-
def mlp(x: torch.Tensor, w: MLPWeights) -> torch.Tensor:
|
40 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
x = gelu_approx(x)
|
42 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
43 |
return x
|
44 |
|
45 |
|
|
|
1 |
+
import torch
|
2 |
+
import torch.nn as nn
|
3 |
+
import torch.nn.functional as F
|
4 |
+
|
5 |
from dataclasses import dataclass
|
6 |
+
from typing import Literal, Optional
|
7 |
|
8 |
+
try:
|
9 |
+
from torchao import quantize_
|
10 |
+
from torchao.quantization import int4_weight_only
|
11 |
+
except ImportError:
|
12 |
+
|
13 |
+
def quantize_(model, quant_mode):
|
14 |
+
raise ImportError(
|
15 |
+
"torchao is not installed. Please install it with `pip install torchao`."
|
16 |
+
)
|
17 |
+
|
18 |
+
def int4_weight_only(group_size):
|
19 |
+
raise ImportError(
|
20 |
+
"torchao is not installed. Please install it with `pip install torchao`."
|
21 |
+
)
|
22 |
|
23 |
|
24 |
def gelu_approx(x):
|
|
|
35 |
return F.linear(x, w.weight, w.bias)
|
36 |
|
37 |
|
38 |
+
def dequantize_tensor(W_q, scale, zero, orig_shape, dtype=torch.bfloat16):
|
39 |
+
_step = W_q.shape[0]
|
40 |
+
W_r = torch.empty([2 * _step, W_q.shape[1]], dtype=dtype, device=W_q.device)
|
41 |
+
W_r[:_step] = (W_q & 0b11110000) >> 4
|
42 |
+
W_r[_step:] = W_q & 0b00001111
|
43 |
+
W_r.sub_(zero).mul_(scale)
|
44 |
+
return W_r.reshape(orig_shape)
|
45 |
+
|
46 |
+
|
47 |
+
class QuantizedLinear(nn.Module):
|
48 |
+
def __init__(
|
49 |
+
self,
|
50 |
+
in_features: int,
|
51 |
+
out_features: int,
|
52 |
+
dtype: torch.dtype,
|
53 |
+
):
|
54 |
+
# TODO: Take group_size as an input instead of hardcoding it here.
|
55 |
+
super().__init__()
|
56 |
+
self.in_features = in_features
|
57 |
+
self.out_features = out_features
|
58 |
+
self.weight = nn.ParameterDict(
|
59 |
+
{
|
60 |
+
"packed": nn.Parameter(
|
61 |
+
torch.empty(
|
62 |
+
out_features * in_features // (128 * 2), 128, dtype=torch.uint8
|
63 |
+
),
|
64 |
+
requires_grad=False,
|
65 |
+
),
|
66 |
+
"scale": nn.Parameter(
|
67 |
+
torch.empty(out_features * in_features // 128, 1),
|
68 |
+
requires_grad=False,
|
69 |
+
),
|
70 |
+
"zero_point": nn.Parameter(
|
71 |
+
torch.empty(out_features * in_features // 128, 1),
|
72 |
+
requires_grad=False,
|
73 |
+
),
|
74 |
+
}
|
75 |
+
)
|
76 |
+
self.bias = nn.Parameter(torch.empty(out_features), requires_grad=False)
|
77 |
+
self.unpacked = False
|
78 |
+
|
79 |
+
def unpack(self):
|
80 |
+
if self.unpacked:
|
81 |
+
return
|
82 |
+
|
83 |
+
self.weight = nn.Parameter(
|
84 |
+
dequantize_tensor(
|
85 |
+
self.weight["packed"],
|
86 |
+
self.weight["scale"],
|
87 |
+
self.weight["zero_point"],
|
88 |
+
(self.out_features, self.in_features),
|
89 |
+
torch.bfloat16,
|
90 |
+
)
|
91 |
+
)
|
92 |
+
with torch.device("meta"):
|
93 |
+
self.linear = nn.Linear(
|
94 |
+
self.in_features, self.out_features, dtype=torch.bfloat16
|
95 |
+
)
|
96 |
+
self.linear.weight = self.weight
|
97 |
+
self.linear.bias = nn.Parameter(
|
98 |
+
self.bias.to(torch.bfloat16), requires_grad=False
|
99 |
+
)
|
100 |
+
|
101 |
+
del self.weight, self.bias
|
102 |
+
quantize_(self, int4_weight_only(group_size=128))
|
103 |
+
self.unpacked = True
|
104 |
+
torch.cuda.empty_cache()
|
105 |
+
|
106 |
+
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
107 |
+
if not self.unpacked:
|
108 |
+
self.unpack()
|
109 |
+
return self.linear(x)
|
110 |
+
|
111 |
+
|
112 |
@dataclass
|
113 |
class LayerNormWeights:
|
114 |
weight: torch.Tensor
|
|
|
126 |
act: Literal["gelu_approx"] = "gelu_approx"
|
127 |
|
128 |
|
129 |
+
def mlp(x: torch.Tensor, w: MLPWeights, lora: Optional[dict] = None) -> torch.Tensor:
|
130 |
+
x0 = w.fc1(x)
|
131 |
+
if lora is not None:
|
132 |
+
x1 = F.linear(F.linear(x, lora["fc1"]["A"]), lora["fc1"]["B"])
|
133 |
+
x = x0 + x1
|
134 |
+
else:
|
135 |
+
x = x0
|
136 |
+
|
137 |
x = gelu_approx(x)
|
138 |
+
|
139 |
+
x0 = w.fc2(x)
|
140 |
+
if lora is not None:
|
141 |
+
x1 = F.linear(F.linear(x, lora["fc2"]["A"]), lora["fc2"]["B"])
|
142 |
+
x = x0 + x1
|
143 |
+
else:
|
144 |
+
x = x0
|
145 |
+
|
146 |
return x
|
147 |
|
148 |
|
lora.py
ADDED
@@ -0,0 +1,82 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import functools
|
2 |
+
import os
|
3 |
+
import shutil
|
4 |
+
import torch
|
5 |
+
|
6 |
+
from pathlib import Path
|
7 |
+
from urllib.request import Request, urlopen
|
8 |
+
from typing import Optional
|
9 |
+
|
10 |
+
|
11 |
+
def variant_cache_dir():
|
12 |
+
hf_hub_cache = os.environ.get("HF_HUB_CACHE")
|
13 |
+
if hf_hub_cache is not None:
|
14 |
+
return Path(hf_hub_cache) / "md_variants"
|
15 |
+
|
16 |
+
hf_home = os.environ.get("HF_HOME")
|
17 |
+
if hf_home is not None:
|
18 |
+
return Path(hf_home) / "hub" / "md_variants"
|
19 |
+
|
20 |
+
return Path("~/.cache/huggingface/hub").expanduser() / "md_variants"
|
21 |
+
|
22 |
+
|
23 |
+
def cached_variant_path(variant_id: str):
|
24 |
+
variant, *rest = variant_id.split("/", 1)
|
25 |
+
step = rest[0] if rest else "final"
|
26 |
+
|
27 |
+
cache_dir = variant_cache_dir() / variant
|
28 |
+
os.makedirs(cache_dir, exist_ok=True)
|
29 |
+
dest = cache_dir / f"{step}.pt"
|
30 |
+
if dest.exists():
|
31 |
+
return dest
|
32 |
+
|
33 |
+
md_endpoint = os.getenv("MOONDREAM_ENDPOINT", "https://api.moondream.ai")
|
34 |
+
|
35 |
+
headers = {"User-Agent": "moondream-torch"}
|
36 |
+
api_key = os.getenv("MOONDREAM_API_KEY")
|
37 |
+
if api_key is not None:
|
38 |
+
headers["X-Moondream-Auth"] = api_key
|
39 |
+
|
40 |
+
req = Request(f"{md_endpoint}/v1/variants/{variant_id}/download", headers=headers)
|
41 |
+
with urlopen(req) as r, open(dest, "wb") as f:
|
42 |
+
shutil.copyfileobj(r, f)
|
43 |
+
return dest
|
44 |
+
|
45 |
+
|
46 |
+
def nest(flat):
|
47 |
+
tree = {}
|
48 |
+
for k, v in flat.items():
|
49 |
+
parts = k.split(".")
|
50 |
+
d = tree
|
51 |
+
for p in parts[:-1]:
|
52 |
+
d = d.setdefault(p, {})
|
53 |
+
d[parts[-1]] = v
|
54 |
+
return tree
|
55 |
+
|
56 |
+
|
57 |
+
@functools.lru_cache(maxsize=5)
|
58 |
+
def variant_state_dict(variant_id: Optional[str] = None, device: str = "cpu"):
|
59 |
+
if variant_id is None:
|
60 |
+
return None
|
61 |
+
|
62 |
+
state_dict = torch.load(
|
63 |
+
cached_variant_path(variant_id), map_location=device, weights_only=True
|
64 |
+
)
|
65 |
+
|
66 |
+
# TODO: Move these into the training code that saves checkpoints...
|
67 |
+
rename_rules = [
|
68 |
+
("text_model.transformer.h", "text.blocks"),
|
69 |
+
(".mixer", ".attn"),
|
70 |
+
(".out_proj", ".proj"),
|
71 |
+
(".Wqkv", ".qkv"),
|
72 |
+
(".parametrizations.weight.0", ""),
|
73 |
+
]
|
74 |
+
new_state_dict = {}
|
75 |
+
for key, tensor in state_dict.items():
|
76 |
+
new_key = key
|
77 |
+
for old, new in rename_rules:
|
78 |
+
if old in new_key:
|
79 |
+
new_key = new_key.replace(old, new)
|
80 |
+
new_state_dict[new_key] = tensor
|
81 |
+
|
82 |
+
return nest(new_state_dict)
|
model.safetensors
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:70a7d94c0c8349eb58ed2d9e636ef2d0916960f321ecabeac6354b8ba3d7403f
|
3 |
+
size 3854538968
|
moondream.py
CHANGED
@@ -11,9 +11,23 @@ from .config import MoondreamConfig
|
|
11 |
from .image_crops import reconstruct_from_crops
|
12 |
from .vision import vision_encoder, vision_projection, prepare_crops, build_vision_model
|
13 |
from .text import build_text_model, text_encoder, lm_head, text_decoder
|
14 |
-
from .region import
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
from .utils import remove_outlier_points
|
16 |
|
|
|
|
|
|
|
|
|
|
|
17 |
|
18 |
TextSamplingSettings = TypedDict(
|
19 |
"TextSamplingSettings",
|
@@ -21,16 +35,18 @@ TextSamplingSettings = TypedDict(
|
|
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
|
@@ -63,43 +79,47 @@ class KVCache(nn.Module):
|
|
63 |
|
64 |
|
65 |
class MoondreamModel(nn.Module):
|
66 |
-
|
|
|
|
|
|
|
67 |
super().__init__()
|
68 |
self.config = config
|
69 |
|
70 |
-
self.tokenizer = Tokenizer.from_pretrained(
|
71 |
-
"vikhyatk/moondream2", revision="2025-01-09"
|
72 |
-
)
|
73 |
self.vision = build_vision_model(config.vision, dtype)
|
74 |
self.text = build_text_model(config.text, dtype)
|
75 |
|
76 |
# Region Model
|
|
|
|
|
|
|
77 |
self.region = nn.ModuleDict(
|
78 |
{
|
79 |
-
"coord_encoder":
|
80 |
config.region.coord_feat_dim, config.region.dim, dtype=dtype
|
81 |
),
|
82 |
"coord_decoder": nn.ModuleDict(
|
83 |
{
|
84 |
-
"fc1":
|
85 |
config.region.dim, config.region.inner_dim, dtype=dtype
|
86 |
),
|
87 |
-
"fc2":
|
88 |
config.region.inner_dim,
|
89 |
config.region.coord_out_dim,
|
90 |
dtype=dtype,
|
91 |
),
|
92 |
}
|
93 |
),
|
94 |
-
"size_encoder":
|
95 |
config.region.size_feat_dim, config.region.dim, dtype=dtype
|
96 |
),
|
97 |
"size_decoder": nn.ModuleDict(
|
98 |
{
|
99 |
-
"fc1":
|
100 |
config.region.dim, config.region.inner_dim, dtype=dtype
|
101 |
),
|
102 |
-
"fc2":
|
103 |
config.region.inner_dim,
|
104 |
config.region.size_out_dim,
|
105 |
dtype=dtype,
|
@@ -151,17 +171,31 @@ class MoondreamModel(nn.Module):
|
|
151 |
def _vis_proj(self, g: torch.Tensor, r: torch.Tensor):
|
152 |
return vision_projection(g, r, self.vision, self.config.vision)
|
153 |
|
154 |
-
def _prefill(
|
155 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
156 |
|
157 |
def _decode_one_tok(
|
158 |
-
self,
|
|
|
|
|
|
|
|
|
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 |
|
164 |
def compile(self):
|
|
|
|
|
|
|
|
|
165 |
# TODO: vision_projection is not being compiled
|
166 |
self._vis_enc = torch.compile(self._vis_enc, fullgraph=True)
|
167 |
self._prefill = torch.compile(self._prefill, fullgraph=True)
|
@@ -171,6 +205,7 @@ class MoondreamModel(nn.Module):
|
|
171 |
|
172 |
def _run_vision_encoder(self, image: Image.Image) -> torch.Tensor:
|
173 |
all_crops, tiling = prepare_crops(image, self.config.vision, device=self.device)
|
|
|
174 |
torch._dynamo.mark_dynamic(all_crops, 0)
|
175 |
|
176 |
outputs = self._vis_enc(all_crops)
|
@@ -192,12 +227,22 @@ class MoondreamModel(nn.Module):
|
|
192 |
|
193 |
return self._vis_proj(global_features, reconstructed)
|
194 |
|
195 |
-
def encode_image(
|
|
|
|
|
|
|
|
|
196 |
if isinstance(image, EncodedImage):
|
197 |
return image
|
198 |
elif not isinstance(image, Image.Image):
|
199 |
raise ValueError("image must be a PIL Image or EncodedImage")
|
200 |
|
|
|
|
|
|
|
|
|
|
|
|
|
201 |
# Run through text model in addition to the vision encoder, to minimize
|
202 |
# re-computation if multiple queries are performed on this image.
|
203 |
with torch.inference_mode():
|
@@ -209,7 +254,7 @@ class MoondreamModel(nn.Module):
|
|
209 |
inputs_embeds = torch.cat([bos_emb, img_emb[None]], dim=1)
|
210 |
mask = self.attn_mask[:, :, 0 : inputs_embeds.size(1), :]
|
211 |
pos_ids = torch.arange(inputs_embeds.size(1), dtype=torch.long)
|
212 |
-
self._prefill(inputs_embeds, mask, pos_ids)
|
213 |
|
214 |
return EncodedImage(
|
215 |
pos=inputs_embeds.size(1),
|
@@ -233,31 +278,167 @@ class MoondreamModel(nn.Module):
|
|
233 |
return next_probs
|
234 |
|
235 |
def _prefill_prompt(
|
236 |
-
self,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
237 |
):
|
238 |
with torch.inference_mode():
|
239 |
prompt_emb = text_encoder(prompt_tokens, self.text)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
240 |
torch._dynamo.mark_dynamic(prompt_emb, 1)
|
241 |
-
|
|
|
|
|
|
|
|
|
242 |
pos_ids = torch.arange(pos, pos + prompt_emb.size(1), dtype=torch.long)
|
243 |
-
|
244 |
-
|
245 |
|
246 |
if temperature == 0:
|
247 |
-
next_token = torch.argmax(
|
248 |
else:
|
249 |
-
probs = torch.softmax(
|
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
|
255 |
|
256 |
-
def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
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)
|
@@ -270,9 +451,21 @@ class MoondreamModel(nn.Module):
|
|
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,
|
|
|
|
|
|
|
|
|
|
|
|
|
276 |
)
|
277 |
|
278 |
def generator(next_token, pos):
|
@@ -287,7 +480,7 @@ class MoondreamModel(nn.Module):
|
|
287 |
|
288 |
while (
|
289 |
next_token_id := next_token.item()
|
290 |
-
) !=
|
291 |
# Add token to our cache
|
292 |
token_cache.append(next_token_id)
|
293 |
|
@@ -307,7 +500,7 @@ class MoondreamModel(nn.Module):
|
|
307 |
print_len += len(printable_text)
|
308 |
if printable_text:
|
309 |
yield printable_text
|
310 |
-
# Otherwise, only
|
311 |
else:
|
312 |
last_space_idx = text.rfind(" ", print_len)
|
313 |
if last_space_idx >= print_len:
|
@@ -319,13 +512,18 @@ class MoondreamModel(nn.Module):
|
|
319 |
with torch.inference_mode():
|
320 |
next_emb = text_encoder(next_token, self.text)
|
321 |
mask[:, :, pos], pos_ids[0] = 1, pos
|
322 |
-
|
|
|
|
|
|
|
323 |
pos += 1
|
324 |
|
325 |
if temperature == 0:
|
326 |
-
next_token = torch.argmax(
|
|
|
|
|
327 |
else:
|
328 |
-
probs = torch.softmax(
|
329 |
probs = self._apply_top_p(probs, top_p)
|
330 |
next_token = torch.multinomial(probs, num_samples=1) # (1, 1)
|
331 |
|
@@ -342,34 +540,82 @@ class MoondreamModel(nn.Module):
|
|
342 |
|
343 |
def query(
|
344 |
self,
|
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.")
|
352 |
|
353 |
-
|
354 |
-
|
355 |
|
356 |
-
|
357 |
-
|
358 |
-
|
359 |
-
|
360 |
-
|
361 |
-
|
362 |
-
|
363 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
364 |
|
365 |
def generator():
|
366 |
-
for token in self.
|
|
|
|
|
367 |
yield token
|
368 |
|
369 |
if stream:
|
370 |
-
return {"answer": generator()}
|
371 |
else:
|
372 |
-
return {"answer": "".join(list(generator()))}
|
373 |
|
374 |
def load_encoded_image(self, encoded_image: EncodedImage):
|
375 |
for b, (k, v) in zip(self.text.blocks, encoded_image.caches):
|
@@ -388,7 +634,7 @@ class MoondreamModel(nn.Module):
|
|
388 |
if length not in self.config.tokenizer.templates["caption"]:
|
389 |
raise ValueError(f"Model does not support caption length '{length}'.")
|
390 |
|
391 |
-
image = self.encode_image(image)
|
392 |
self.load_encoded_image(image)
|
393 |
|
394 |
prompt_tokens = torch.tensor(
|
@@ -396,7 +642,7 @@ class MoondreamModel(nn.Module):
|
|
396 |
)
|
397 |
|
398 |
def generator():
|
399 |
-
for token in self.
|
400 |
yield token
|
401 |
|
402 |
if stream:
|
@@ -411,6 +657,7 @@ class MoondreamModel(nn.Module):
|
|
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)
|
@@ -430,7 +677,7 @@ class MoondreamModel(nn.Module):
|
|
430 |
|
431 |
# Decode y-coordinate
|
432 |
mask[:, :, pos], pos_ids[0] = 1, pos
|
433 |
-
_, hidden = self._decode_one_tok(next_emb, mask, pos_ids)
|
434 |
pos += 1
|
435 |
y_logits = decode_coordinate(hidden, self.region)
|
436 |
y_center = torch.argmax(y_logits, dim=-1) / y_logits.size(-1)
|
@@ -441,7 +688,7 @@ class MoondreamModel(nn.Module):
|
|
441 |
# Decode size
|
442 |
if include_size:
|
443 |
mask[:, :, pos], pos_ids[0] = 1, pos
|
444 |
-
logits, hidden = self._decode_one_tok(next_emb, mask, pos_ids)
|
445 |
pos += 1
|
446 |
size_logits = decode_size(hidden, self.region)
|
447 |
|
@@ -479,7 +726,7 @@ class MoondreamModel(nn.Module):
|
|
479 |
|
480 |
# Decode next token (x-coordinate, or eos)
|
481 |
mask[:, :, pos], pos_ids[0] = 1, pos
|
482 |
-
logits, hidden = self._decode_one_tok(next_emb, mask, pos_ids)
|
483 |
pos += 1
|
484 |
next_token = torch.argmax(logits, dim=-1)
|
485 |
|
@@ -494,7 +741,7 @@ class MoondreamModel(nn.Module):
|
|
494 |
if self.config.tokenizer.templates["detect"] is None:
|
495 |
raise NotImplementedError("Model does not support object detection.")
|
496 |
|
497 |
-
image = self.encode_image(image)
|
498 |
self.load_encoded_image(image)
|
499 |
|
500 |
prompt_tokens = torch.tensor(
|
@@ -506,8 +753,14 @@ class MoondreamModel(nn.Module):
|
|
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 |
|
@@ -517,7 +770,12 @@ class MoondreamModel(nn.Module):
|
|
517 |
else DEFAULT_MAX_OBJECTS
|
518 |
)
|
519 |
objects = self._generate_points(
|
520 |
-
hidden,
|
|
|
|
|
|
|
|
|
|
|
521 |
)
|
522 |
|
523 |
return {"objects": objects}
|
@@ -531,7 +789,7 @@ class MoondreamModel(nn.Module):
|
|
531 |
if self.config.tokenizer.templates["point"] is None:
|
532 |
raise NotImplementedError("Model does not support pointing.")
|
533 |
|
534 |
-
image = self.encode_image(image)
|
535 |
self.load_encoded_image(image)
|
536 |
|
537 |
prompt_tokens = torch.tensor(
|
@@ -543,8 +801,14 @@ class MoondreamModel(nn.Module):
|
|
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 |
|
@@ -554,7 +818,12 @@ class MoondreamModel(nn.Module):
|
|
554 |
else DEFAULT_MAX_OBJECTS
|
555 |
)
|
556 |
objects = self._generate_points(
|
557 |
-
hidden,
|
|
|
|
|
|
|
|
|
|
|
558 |
)
|
559 |
|
560 |
return {"points": objects}
|
@@ -579,11 +848,11 @@ class MoondreamModel(nn.Module):
|
|
579 |
self.text,
|
580 |
)
|
581 |
x_emb = encode_coordinate(
|
582 |
-
torch.tensor([[[source[0]]]], device=self.device, dtype=torch.
|
583 |
self.region,
|
584 |
)
|
585 |
y_emb = encode_coordinate(
|
586 |
-
torch.tensor([[[source[1]]]], device=self.device, dtype=torch.
|
587 |
self.region,
|
588 |
)
|
589 |
|
@@ -595,7 +864,7 @@ class MoondreamModel(nn.Module):
|
|
595 |
pos_ids = torch.arange(
|
596 |
image.pos, image.pos + prompt_emb.size(1), dtype=torch.long
|
597 |
)
|
598 |
-
hidden = self._prefill(prompt_emb, mask, pos_ids)
|
599 |
logits = lm_head(hidden, self.text)
|
600 |
next_token = torch.argmax(logits, dim=-1)
|
601 |
pos = image.pos + prompt_emb.size(1)
|
|
|
11 |
from .image_crops import reconstruct_from_crops
|
12 |
from .vision import vision_encoder, vision_projection, prepare_crops, build_vision_model
|
13 |
from .text import build_text_model, text_encoder, lm_head, text_decoder
|
14 |
+
from .region import (
|
15 |
+
decode_coordinate,
|
16 |
+
encode_coordinate,
|
17 |
+
decode_size,
|
18 |
+
encode_size,
|
19 |
+
encode_spatial_refs,
|
20 |
+
SpatialRefs,
|
21 |
+
)
|
22 |
+
from .layers import QuantizedLinear
|
23 |
+
from .lora import variant_state_dict
|
24 |
from .utils import remove_outlier_points
|
25 |
|
26 |
+
ImageEncodingSettings = TypedDict(
|
27 |
+
"ImageEncodingSettings",
|
28 |
+
{"variant": str},
|
29 |
+
total=False,
|
30 |
+
)
|
31 |
|
32 |
TextSamplingSettings = TypedDict(
|
33 |
"TextSamplingSettings",
|
|
|
35 |
"max_tokens": int,
|
36 |
"temperature": float,
|
37 |
"top_p": float,
|
38 |
+
"variant": str,
|
39 |
},
|
40 |
total=False,
|
41 |
)
|
42 |
|
43 |
ObjectSamplingSettings = TypedDict(
|
44 |
"ObjectSamplingSettings",
|
45 |
+
{"max_objects": int, "variant": str},
|
46 |
total=False,
|
47 |
)
|
48 |
|
49 |
+
|
50 |
DEFAULT_MAX_TOKENS = 768
|
51 |
DEFAULT_TEMPERATURE = 0.5
|
52 |
DEFAULT_TOP_P = 0.3
|
|
|
79 |
|
80 |
|
81 |
class MoondreamModel(nn.Module):
|
82 |
+
|
83 |
+
def __init__(
|
84 |
+
self, config: MoondreamConfig, dtype=torch.bfloat16, setup_caches=True
|
85 |
+
):
|
86 |
super().__init__()
|
87 |
self.config = config
|
88 |
|
89 |
+
self.tokenizer = Tokenizer.from_pretrained("moondream/starmie-v1")
|
|
|
|
|
90 |
self.vision = build_vision_model(config.vision, dtype)
|
91 |
self.text = build_text_model(config.text, dtype)
|
92 |
|
93 |
# Region Model
|
94 |
+
linear_cls = (
|
95 |
+
QuantizedLinear if config.region.group_size is not None else nn.Linear
|
96 |
+
)
|
97 |
self.region = nn.ModuleDict(
|
98 |
{
|
99 |
+
"coord_encoder": linear_cls(
|
100 |
config.region.coord_feat_dim, config.region.dim, dtype=dtype
|
101 |
),
|
102 |
"coord_decoder": nn.ModuleDict(
|
103 |
{
|
104 |
+
"fc1": linear_cls(
|
105 |
config.region.dim, config.region.inner_dim, dtype=dtype
|
106 |
),
|
107 |
+
"fc2": linear_cls(
|
108 |
config.region.inner_dim,
|
109 |
config.region.coord_out_dim,
|
110 |
dtype=dtype,
|
111 |
),
|
112 |
}
|
113 |
),
|
114 |
+
"size_encoder": linear_cls(
|
115 |
config.region.size_feat_dim, config.region.dim, dtype=dtype
|
116 |
),
|
117 |
"size_decoder": nn.ModuleDict(
|
118 |
{
|
119 |
+
"fc1": linear_cls(
|
120 |
config.region.dim, config.region.inner_dim, dtype=dtype
|
121 |
),
|
122 |
+
"fc2": linear_cls(
|
123 |
config.region.inner_dim,
|
124 |
config.region.size_out_dim,
|
125 |
dtype=dtype,
|
|
|
171 |
def _vis_proj(self, g: torch.Tensor, r: torch.Tensor):
|
172 |
return vision_projection(g, r, self.vision, self.config.vision)
|
173 |
|
174 |
+
def _prefill(
|
175 |
+
self,
|
176 |
+
x: torch.Tensor,
|
177 |
+
attn_mask: torch.Tensor,
|
178 |
+
pos_ids: torch.Tensor,
|
179 |
+
lora: Optional[torch.Tensor],
|
180 |
+
):
|
181 |
+
return text_decoder(x, self.text, attn_mask, pos_ids, self.config.text, lora)
|
182 |
|
183 |
def _decode_one_tok(
|
184 |
+
self,
|
185 |
+
x: torch.Tensor,
|
186 |
+
attn_mask: torch.Tensor,
|
187 |
+
pos_ids: torch.Tensor,
|
188 |
+
lora: Optional[torch.Tensor],
|
189 |
):
|
190 |
+
hidden = text_decoder(x, self.text, attn_mask, pos_ids, self.config.text, lora)
|
191 |
logits = lm_head(hidden, self.text)
|
192 |
return logits, hidden
|
193 |
|
194 |
def compile(self):
|
195 |
+
for module in self.modules():
|
196 |
+
if isinstance(module, QuantizedLinear):
|
197 |
+
module.unpack()
|
198 |
+
|
199 |
# TODO: vision_projection is not being compiled
|
200 |
self._vis_enc = torch.compile(self._vis_enc, fullgraph=True)
|
201 |
self._prefill = torch.compile(self._prefill, fullgraph=True)
|
|
|
205 |
|
206 |
def _run_vision_encoder(self, image: Image.Image) -> torch.Tensor:
|
207 |
all_crops, tiling = prepare_crops(image, self.config.vision, device=self.device)
|
208 |
+
|
209 |
torch._dynamo.mark_dynamic(all_crops, 0)
|
210 |
|
211 |
outputs = self._vis_enc(all_crops)
|
|
|
227 |
|
228 |
return self._vis_proj(global_features, reconstructed)
|
229 |
|
230 |
+
def encode_image(
|
231 |
+
self,
|
232 |
+
image: Union[Image.Image, EncodedImage],
|
233 |
+
settings: Optional[ImageEncodingSettings] = None,
|
234 |
+
) -> EncodedImage:
|
235 |
if isinstance(image, EncodedImage):
|
236 |
return image
|
237 |
elif not isinstance(image, Image.Image):
|
238 |
raise ValueError("image must be a PIL Image or EncodedImage")
|
239 |
|
240 |
+
lora = (
|
241 |
+
variant_state_dict(settings["variant"], device=self.device)
|
242 |
+
if settings is not None and settings["variant"] is not None
|
243 |
+
else None
|
244 |
+
)
|
245 |
+
|
246 |
# Run through text model in addition to the vision encoder, to minimize
|
247 |
# re-computation if multiple queries are performed on this image.
|
248 |
with torch.inference_mode():
|
|
|
254 |
inputs_embeds = torch.cat([bos_emb, img_emb[None]], dim=1)
|
255 |
mask = self.attn_mask[:, :, 0 : inputs_embeds.size(1), :]
|
256 |
pos_ids = torch.arange(inputs_embeds.size(1), dtype=torch.long)
|
257 |
+
self._prefill(inputs_embeds, mask, pos_ids, lora)
|
258 |
|
259 |
return EncodedImage(
|
260 |
pos=inputs_embeds.size(1),
|
|
|
278 |
return next_probs
|
279 |
|
280 |
def _prefill_prompt(
|
281 |
+
self,
|
282 |
+
prompt_tokens: torch.Tensor,
|
283 |
+
pos: int,
|
284 |
+
temperature: float,
|
285 |
+
top_p: float,
|
286 |
+
spatial_refs: Optional[SpatialRefs] = None,
|
287 |
+
attn_mask: Optional[torch.Tensor] = None,
|
288 |
+
lora: Optional[dict] = None,
|
289 |
):
|
290 |
with torch.inference_mode():
|
291 |
prompt_emb = text_encoder(prompt_tokens, self.text)
|
292 |
+
|
293 |
+
if spatial_refs:
|
294 |
+
encoded_refs = encode_spatial_refs(spatial_refs, self.region)
|
295 |
+
prompt_emb[prompt_tokens == self.config.tokenizer.coord_id] = (
|
296 |
+
encoded_refs["coords"]
|
297 |
+
)
|
298 |
+
if encoded_refs["sizes"] is not None:
|
299 |
+
prompt_emb[prompt_tokens == self.config.tokenizer.size_id] = (
|
300 |
+
encoded_refs["sizes"]
|
301 |
+
)
|
302 |
+
|
303 |
torch._dynamo.mark_dynamic(prompt_emb, 1)
|
304 |
+
|
305 |
+
if attn_mask is None:
|
306 |
+
attn_mask = self.attn_mask
|
307 |
+
|
308 |
+
mask = attn_mask[:, :, pos : pos + prompt_emb.size(1), :]
|
309 |
pos_ids = torch.arange(pos, pos + prompt_emb.size(1), dtype=torch.long)
|
310 |
+
hidden_BC = self._prefill(prompt_emb, mask, pos_ids, lora)
|
311 |
+
logits_BV = lm_head(hidden_BC, self.text)
|
312 |
|
313 |
if temperature == 0:
|
314 |
+
next_token = torch.argmax(logits_BV, dim=-1).unsqueeze(1)
|
315 |
else:
|
316 |
+
probs = torch.softmax(logits_BV / temperature, dim=-1)
|
317 |
probs = self._apply_top_p(probs, top_p)
|
318 |
next_token = torch.multinomial(probs, num_samples=1)
|
319 |
|
320 |
pos = pos + prompt_emb.size(1)
|
321 |
+
return logits_BV, hidden_BC, next_token, pos
|
322 |
|
323 |
+
def _generate_reasoning(
|
324 |
+
self,
|
325 |
+
prompt_tokens,
|
326 |
+
pos,
|
327 |
+
settings: Optional[TextSamplingSettings] = None,
|
328 |
+
spatial_refs: Optional[SpatialRefs] = None,
|
329 |
+
attn_mask: Optional[torch.Tensor] = None,
|
330 |
+
) -> Tuple[int, str, List[dict]]:
|
331 |
+
max_tokens = (
|
332 |
+
settings.get("max_tokens", DEFAULT_MAX_TOKENS)
|
333 |
+
if settings
|
334 |
+
else DEFAULT_MAX_TOKENS
|
335 |
+
)
|
336 |
+
temperature = (
|
337 |
+
settings.get("temperature", DEFAULT_TEMPERATURE)
|
338 |
+
if settings
|
339 |
+
else DEFAULT_TEMPERATURE
|
340 |
+
)
|
341 |
+
lora = (
|
342 |
+
variant_state_dict(settings["variant"], device=self.device)
|
343 |
+
if settings is not None and "variant" in settings
|
344 |
+
else None
|
345 |
+
)
|
346 |
+
|
347 |
+
top_p = settings.get("top_p", DEFAULT_TOP_P) if settings else DEFAULT_TOP_P
|
348 |
+
eos_id = self.config.tokenizer.answer_id
|
349 |
+
|
350 |
+
_, last_hidden_BC, next_token, pos = self._prefill_prompt(
|
351 |
+
prompt_tokens,
|
352 |
+
pos,
|
353 |
+
temperature,
|
354 |
+
top_p,
|
355 |
+
spatial_refs,
|
356 |
+
attn_mask=attn_mask,
|
357 |
+
lora=lora,
|
358 |
+
)
|
359 |
+
|
360 |
+
text_token_chunks = [[]]
|
361 |
+
grounding_chunks = [[]]
|
362 |
+
|
363 |
+
mask = torch.zeros(1, 1, 2048, device=self.device, dtype=torch.bool)
|
364 |
+
mask[:, :, :pos] = 1
|
365 |
+
pos_ids = torch.tensor([pos], device=self.device, dtype=torch.long)
|
366 |
+
generated_tokens = 0
|
367 |
+
|
368 |
+
while (
|
369 |
+
next_token_id := next_token.item()
|
370 |
+
) != eos_id and generated_tokens < max_tokens:
|
371 |
+
if (
|
372 |
+
next_token_id == self.config.tokenizer.start_ground_points_id
|
373 |
+
or next_token_id == self.config.tokenizer.end_ground_id
|
374 |
+
):
|
375 |
+
text_token_chunks.append([])
|
376 |
+
grounding_chunks.append([])
|
377 |
+
|
378 |
+
text_token_chunks[-1].append(next_token_id)
|
379 |
+
|
380 |
+
with torch.inference_mode():
|
381 |
+
if next_token_id == self.config.tokenizer.coord_id:
|
382 |
+
coord_logits = decode_coordinate(last_hidden_BC, self.region)
|
383 |
+
coord = torch.argmax(coord_logits, dim=-1) / coord_logits.size(-1)
|
384 |
+
grounding_chunks[-1].append(coord.item())
|
385 |
+
|
386 |
+
next_emb = encode_coordinate(
|
387 |
+
coord.to(dtype=coord_logits.dtype), self.region
|
388 |
+
).unsqueeze(0)
|
389 |
+
else:
|
390 |
+
next_emb = text_encoder(next_token, self.text)
|
391 |
+
|
392 |
+
mask[:, :, pos], pos_ids[0] = 1, pos
|
393 |
+
|
394 |
+
logits_BV, last_hidden_BC = self._decode_one_tok(
|
395 |
+
next_emb, mask, pos_ids, lora
|
396 |
+
)
|
397 |
+
logits_BV[:, self.config.tokenizer.eos_id] = float("-inf")
|
398 |
+
logits_BV[:, self.config.tokenizer.size_id] = float("-inf")
|
399 |
+
|
400 |
+
pos += 1
|
401 |
+
|
402 |
+
if temperature == 0:
|
403 |
+
next_token = torch.argmax(logits_BV, dim=-1).unsqueeze(1) # (1, 1)
|
404 |
+
else:
|
405 |
+
probs = torch.softmax(logits_BV / temperature, dim=-1) # (1, V)
|
406 |
+
probs = self._apply_top_p(probs, top_p)
|
407 |
+
next_token = torch.multinomial(probs, num_samples=1) # (1, 1)
|
408 |
+
|
409 |
+
generated_tokens += 1
|
410 |
+
|
411 |
+
text_chunks = [
|
412 |
+
self.tokenizer.decode(chunk_tokens) for chunk_tokens in text_token_chunks
|
413 |
+
]
|
414 |
+
text = "".join(text_chunks)
|
415 |
+
|
416 |
+
start_idx = 0
|
417 |
+
grounding = []
|
418 |
+
for text_chunk, grounding_chunk in zip(text_chunks, grounding_chunks):
|
419 |
+
if len(grounding_chunk) > 1:
|
420 |
+
points = []
|
421 |
+
for i in range(0, len(grounding_chunk) - (len(grounding_chunk) % 2), 2):
|
422 |
+
points.append((grounding_chunk[i], grounding_chunk[i + 1]))
|
423 |
+
grounding.append(
|
424 |
+
{
|
425 |
+
"start_idx": start_idx,
|
426 |
+
"end_idx": start_idx + len(text_chunk),
|
427 |
+
"points": points,
|
428 |
+
}
|
429 |
+
)
|
430 |
+
start_idx += len(text_chunk)
|
431 |
+
|
432 |
+
return pos, text, grounding
|
433 |
+
|
434 |
+
def _generate_answer(
|
435 |
self,
|
436 |
prompt_tokens: torch.Tensor,
|
437 |
pos: int,
|
438 |
settings: Optional[TextSamplingSettings] = None,
|
439 |
+
spatial_refs: Optional[SpatialRefs] = None,
|
440 |
+
eos_id: Optional[int] = None,
|
441 |
+
attn_mask: Optional[torch.Tensor] = None,
|
442 |
):
|
443 |
max_tokens = (
|
444 |
settings.get("max_tokens", DEFAULT_MAX_TOKENS)
|
|
|
451 |
else DEFAULT_TEMPERATURE
|
452 |
)
|
453 |
top_p = settings.get("top_p", DEFAULT_TOP_P) if settings else DEFAULT_TOP_P
|
454 |
+
eos_id = eos_id if eos_id is not None else self.config.tokenizer.eos_id
|
455 |
+
lora = (
|
456 |
+
variant_state_dict(settings["variant"], device=self.device)
|
457 |
+
if settings is not None and "variant" in settings
|
458 |
+
else None
|
459 |
+
)
|
460 |
|
461 |
_, _, next_token, pos = self._prefill_prompt(
|
462 |
+
prompt_tokens,
|
463 |
+
pos,
|
464 |
+
temperature,
|
465 |
+
top_p,
|
466 |
+
spatial_refs,
|
467 |
+
attn_mask=attn_mask,
|
468 |
+
lora=lora,
|
469 |
)
|
470 |
|
471 |
def generator(next_token, pos):
|
|
|
480 |
|
481 |
while (
|
482 |
next_token_id := next_token.item()
|
483 |
+
) != eos_id and generated_tokens < max_tokens:
|
484 |
# Add token to our cache
|
485 |
token_cache.append(next_token_id)
|
486 |
|
|
|
500 |
print_len += len(printable_text)
|
501 |
if printable_text:
|
502 |
yield printable_text
|
503 |
+
# Otherwise, only yield up to the last space to avoid cutting words
|
504 |
else:
|
505 |
last_space_idx = text.rfind(" ", print_len)
|
506 |
if last_space_idx >= print_len:
|
|
|
512 |
with torch.inference_mode():
|
513 |
next_emb = text_encoder(next_token, self.text)
|
514 |
mask[:, :, pos], pos_ids[0] = 1, pos
|
515 |
+
|
516 |
+
logits_BV, _ = self._decode_one_tok(next_emb, mask, pos_ids, lora)
|
517 |
+
logits_BV[:, self.config.tokenizer.answer_id] = float("-inf")
|
518 |
+
|
519 |
pos += 1
|
520 |
|
521 |
if temperature == 0:
|
522 |
+
next_token = torch.argmax(logits_BV, dim=-1).unsqueeze(
|
523 |
+
1
|
524 |
+
) # (1, 1)
|
525 |
else:
|
526 |
+
probs = torch.softmax(logits_BV / temperature, dim=-1) # (1, V)
|
527 |
probs = self._apply_top_p(probs, top_p)
|
528 |
next_token = torch.multinomial(probs, num_samples=1) # (1, 1)
|
529 |
|
|
|
540 |
|
541 |
def query(
|
542 |
self,
|
543 |
+
image: Optional[Union[Image.Image, EncodedImage]] = None,
|
544 |
+
question: str = None,
|
545 |
+
reasoning: bool = False,
|
546 |
+
spatial_refs: Optional[SpatialRefs] = None,
|
547 |
stream: bool = False,
|
548 |
settings: Optional[TextSamplingSettings] = None,
|
549 |
):
|
550 |
if self.config.tokenizer.templates["query"] is None:
|
551 |
raise NotImplementedError("Model does not support querying.")
|
552 |
|
553 |
+
if question is None:
|
554 |
+
raise ValueError("question must be provided.")
|
555 |
|
556 |
+
if spatial_refs and image is None:
|
557 |
+
raise ValueError("spatial_refs can only be used with an image.")
|
558 |
+
|
559 |
+
attn_mask = self.attn_mask
|
560 |
+
if image is not None:
|
561 |
+
image = self.encode_image(image, settings)
|
562 |
+
self.load_encoded_image(image)
|
563 |
+
pos = image.pos
|
564 |
+
prompt_toks = self.config.tokenizer.templates["query"]["prefix"]
|
565 |
+
else:
|
566 |
+
self._setup_caches()
|
567 |
+
pos = 0
|
568 |
+
prompt_toks = [
|
569 |
+
self.config.tokenizer.bos_id
|
570 |
+
] + self.config.tokenizer.templates["query"]["prefix"]
|
571 |
+
max_context = self.config.text.max_context
|
572 |
+
attn_mask = torch.tril(
|
573 |
+
torch.ones(1, 1, max_context, max_context, dtype=torch.bool)
|
574 |
+
).to(self.device)
|
575 |
+
|
576 |
+
spatial_toks = []
|
577 |
+
if spatial_refs:
|
578 |
+
for ref in spatial_refs:
|
579 |
+
coord_id = self.config.tokenizer.coord_id
|
580 |
+
size_id = self.config.tokenizer.size_id
|
581 |
+
if len(ref) == 2:
|
582 |
+
spatial_toks.extend([coord_id, coord_id])
|
583 |
+
else:
|
584 |
+
spatial_toks.extend([coord_id, coord_id, size_id])
|
585 |
+
|
586 |
+
prompt_tokens = [
|
587 |
+
prompt_toks
|
588 |
+
+ spatial_toks
|
589 |
+
+ self.tokenizer.encode(question).ids
|
590 |
+
+ self.config.tokenizer.templates["query"]["suffix"]
|
591 |
+
]
|
592 |
+
|
593 |
+
if reasoning:
|
594 |
+
prompt_tokens[0] += [self.config.tokenizer.thinking_id]
|
595 |
+
prompt_tokens = torch.tensor(prompt_tokens, device=self.device)
|
596 |
+
pos, reasoning_text, reasoning_grounding = self._generate_reasoning(
|
597 |
+
prompt_tokens, pos, settings, spatial_refs, attn_mask=attn_mask
|
598 |
+
)
|
599 |
+
prompt_tokens = [self.config.tokenizer.templates["query"]["suffix"]]
|
600 |
+
reasoning_dict = {
|
601 |
+
"reasoning": {"text": reasoning_text, "grounding": reasoning_grounding}
|
602 |
+
}
|
603 |
+
else:
|
604 |
+
prompt_tokens[0] += self.config.tokenizer.templates["query"]["suffix"]
|
605 |
+
reasoning_dict = {}
|
606 |
+
|
607 |
+
prompt_tokens = torch.tensor(prompt_tokens, device=self.device)
|
608 |
|
609 |
def generator():
|
610 |
+
for token in self._generate_answer(
|
611 |
+
prompt_tokens, pos, settings, spatial_refs, attn_mask=attn_mask
|
612 |
+
):
|
613 |
yield token
|
614 |
|
615 |
if stream:
|
616 |
+
return {**reasoning_dict, "answer": generator()}
|
617 |
else:
|
618 |
+
return {**reasoning_dict, "answer": "".join(list(generator()))}
|
619 |
|
620 |
def load_encoded_image(self, encoded_image: EncodedImage):
|
621 |
for b, (k, v) in zip(self.text.blocks, encoded_image.caches):
|
|
|
634 |
if length not in self.config.tokenizer.templates["caption"]:
|
635 |
raise ValueError(f"Model does not support caption length '{length}'.")
|
636 |
|
637 |
+
image = self.encode_image(image, settings)
|
638 |
self.load_encoded_image(image)
|
639 |
|
640 |
prompt_tokens = torch.tensor(
|
|
|
642 |
)
|
643 |
|
644 |
def generator():
|
645 |
+
for token in self._generate_answer(prompt_tokens, image.pos, settings):
|
646 |
yield token
|
647 |
|
648 |
if stream:
|
|
|
657 |
pos: int,
|
658 |
include_size: bool = True,
|
659 |
max_objects: int = DEFAULT_MAX_OBJECTS,
|
660 |
+
lora: Optional[dict] = None,
|
661 |
):
|
662 |
out = []
|
663 |
mask = torch.zeros(1, 1, 2048, device=self.device, dtype=torch.bool)
|
|
|
677 |
|
678 |
# Decode y-coordinate
|
679 |
mask[:, :, pos], pos_ids[0] = 1, pos
|
680 |
+
_, hidden = self._decode_one_tok(next_emb, mask, pos_ids, lora)
|
681 |
pos += 1
|
682 |
y_logits = decode_coordinate(hidden, self.region)
|
683 |
y_center = torch.argmax(y_logits, dim=-1) / y_logits.size(-1)
|
|
|
688 |
# Decode size
|
689 |
if include_size:
|
690 |
mask[:, :, pos], pos_ids[0] = 1, pos
|
691 |
+
logits, hidden = self._decode_one_tok(next_emb, mask, pos_ids, lora)
|
692 |
pos += 1
|
693 |
size_logits = decode_size(hidden, self.region)
|
694 |
|
|
|
726 |
|
727 |
# Decode next token (x-coordinate, or eos)
|
728 |
mask[:, :, pos], pos_ids[0] = 1, pos
|
729 |
+
logits, hidden = self._decode_one_tok(next_emb, mask, pos_ids, lora)
|
730 |
pos += 1
|
731 |
next_token = torch.argmax(logits, dim=-1)
|
732 |
|
|
|
741 |
if self.config.tokenizer.templates["detect"] is None:
|
742 |
raise NotImplementedError("Model does not support object detection.")
|
743 |
|
744 |
+
image = self.encode_image(image, settings)
|
745 |
self.load_encoded_image(image)
|
746 |
|
747 |
prompt_tokens = torch.tensor(
|
|
|
753 |
device=self.device,
|
754 |
)
|
755 |
|
756 |
+
lora = (
|
757 |
+
variant_state_dict(settings["variant"], device=self.device)
|
758 |
+
if settings is not None and "variant" in settings
|
759 |
+
else None
|
760 |
+
)
|
761 |
+
|
762 |
_, hidden, next_token, pos = self._prefill_prompt(
|
763 |
+
prompt_tokens, image.pos, temperature=0, top_p=0, lora=lora
|
764 |
)
|
765 |
hidden = hidden[:, -1:, :]
|
766 |
|
|
|
770 |
else DEFAULT_MAX_OBJECTS
|
771 |
)
|
772 |
objects = self._generate_points(
|
773 |
+
hidden,
|
774 |
+
next_token,
|
775 |
+
pos,
|
776 |
+
include_size=True,
|
777 |
+
max_objects=max_objects,
|
778 |
+
lora=lora,
|
779 |
)
|
780 |
|
781 |
return {"objects": objects}
|
|
|
789 |
if self.config.tokenizer.templates["point"] is None:
|
790 |
raise NotImplementedError("Model does not support pointing.")
|
791 |
|
792 |
+
image = self.encode_image(image, settings)
|
793 |
self.load_encoded_image(image)
|
794 |
|
795 |
prompt_tokens = torch.tensor(
|
|
|
801 |
device=self.device,
|
802 |
)
|
803 |
|
804 |
+
lora = (
|
805 |
+
variant_state_dict(settings["variant"], device=self.device)
|
806 |
+
if settings is not None and "variant" in settings
|
807 |
+
else None
|
808 |
+
)
|
809 |
+
|
810 |
_, hidden, next_token, pos = self._prefill_prompt(
|
811 |
+
prompt_tokens, image.pos, temperature=0, top_p=0, lora=lora
|
812 |
)
|
813 |
hidden = hidden[:, -1:, :]
|
814 |
|
|
|
818 |
else DEFAULT_MAX_OBJECTS
|
819 |
)
|
820 |
objects = self._generate_points(
|
821 |
+
hidden,
|
822 |
+
next_token,
|
823 |
+
pos,
|
824 |
+
include_size=False,
|
825 |
+
max_objects=max_objects,
|
826 |
+
lora=lora,
|
827 |
)
|
828 |
|
829 |
return {"points": objects}
|
|
|
848 |
self.text,
|
849 |
)
|
850 |
x_emb = encode_coordinate(
|
851 |
+
torch.tensor([[[source[0]]]], device=self.device, dtype=torch.bfloat16),
|
852 |
self.region,
|
853 |
)
|
854 |
y_emb = encode_coordinate(
|
855 |
+
torch.tensor([[[source[1]]]], device=self.device, dtype=torch.bfloat16),
|
856 |
self.region,
|
857 |
)
|
858 |
|
|
|
864 |
pos_ids = torch.arange(
|
865 |
image.pos, image.pos + prompt_emb.size(1), dtype=torch.long
|
866 |
)
|
867 |
+
hidden = self._prefill(prompt_emb, mask, pos_ids, lora=None)
|
868 |
logits = lm_head(hidden, self.text)
|
869 |
next_token = torch.argmax(logits, dim=-1)
|
870 |
pos = image.pos + prompt_emb.size(1)
|
region.py
CHANGED
@@ -2,7 +2,11 @@ import torch
|
|
2 |
import torch.nn as nn
|
3 |
import math
|
4 |
|
5 |
-
from
|
|
|
|
|
|
|
|
|
6 |
|
7 |
|
8 |
def fourier_features(x: torch.Tensor, w: torch.Tensor) -> torch.Tensor:
|
@@ -36,7 +40,7 @@ def encode_coordinate(coord: torch.Tensor, w: nn.Module) -> torch.Tensor:
|
|
36 |
Returns:
|
37 |
Encoded hidden states tensor for input to text model
|
38 |
"""
|
39 |
-
return
|
40 |
|
41 |
|
42 |
def decode_coordinate(hidden_state: torch.Tensor, w: nn.Module) -> torch.Tensor:
|
@@ -64,7 +68,7 @@ def encode_size(size: torch.Tensor, w: nn.Module) -> torch.Tensor:
|
|
64 |
Returns:
|
65 |
Encoded hidden states tensor for input to text model
|
66 |
"""
|
67 |
-
return
|
68 |
|
69 |
|
70 |
def decode_size(hidden_state: torch.Tensor, w: nn.Module) -> torch.Tensor:
|
@@ -87,3 +91,46 @@ def decode_size(hidden_state: torch.Tensor, w: nn.Module) -> torch.Tensor:
|
|
87 |
Shape is (2, 1024) where the first dimension corresponds to width and height.
|
88 |
"""
|
89 |
return mlp(hidden_state, w.size_decoder).view(2, -1)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
import torch.nn as nn
|
3 |
import math
|
4 |
|
5 |
+
from typing import List, Tuple, Union
|
6 |
+
|
7 |
+
from .layers import mlp
|
8 |
+
|
9 |
+
SpatialRefs = List[Union[Tuple[float, float], Tuple[float, float, float, float]]]
|
10 |
|
11 |
|
12 |
def fourier_features(x: torch.Tensor, w: torch.Tensor) -> torch.Tensor:
|
|
|
40 |
Returns:
|
41 |
Encoded hidden states tensor for input to text model
|
42 |
"""
|
43 |
+
return w.coord_encoder(fourier_features(coord, w.coord_features))
|
44 |
|
45 |
|
46 |
def decode_coordinate(hidden_state: torch.Tensor, w: nn.Module) -> torch.Tensor:
|
|
|
68 |
Returns:
|
69 |
Encoded hidden states tensor for input to text model
|
70 |
"""
|
71 |
+
return w.size_encoder(fourier_features(size, w.size_features))
|
72 |
|
73 |
|
74 |
def decode_size(hidden_state: torch.Tensor, w: nn.Module) -> torch.Tensor:
|
|
|
91 |
Shape is (2, 1024) where the first dimension corresponds to width and height.
|
92 |
"""
|
93 |
return mlp(hidden_state, w.size_decoder).view(2, -1)
|
94 |
+
|
95 |
+
|
96 |
+
def encode_spatial_refs(spatial_refs: SpatialRefs, w: nn.Module) -> torch.Tensor:
|
97 |
+
"""
|
98 |
+
Takes a list of spatial references (points or regions) and encodes them into
|
99 |
+
hidden states for input to the text model.
|
100 |
+
|
101 |
+
Args:
|
102 |
+
spatial_refs: List of spatial references (points or boxes)
|
103 |
+
- Points are represented as normalized (x, y) tuples
|
104 |
+
- Boxes are represented as normalized (x_min, y_min, x_max, y_max) tuples
|
105 |
+
|
106 |
+
Returns:
|
107 |
+
{"coords": torch.Tensor, "sizes": Optional[torch.Tensor]}
|
108 |
+
"""
|
109 |
+
coords, sizes = [], []
|
110 |
+
for ref in spatial_refs:
|
111 |
+
if len(ref) == 2:
|
112 |
+
coords.append(ref[0])
|
113 |
+
coords.append(ref[1])
|
114 |
+
else:
|
115 |
+
x_c = (ref[0] + ref[2]) / 2
|
116 |
+
y_c = (ref[1] + ref[3]) / 2
|
117 |
+
width = ref[2] - ref[0]
|
118 |
+
height = ref[3] - ref[1]
|
119 |
+
coords.append(x_c)
|
120 |
+
coords.append(y_c)
|
121 |
+
sizes.append([width, height])
|
122 |
+
|
123 |
+
coords = torch.tensor(
|
124 |
+
coords, device=w.coord_features.device, dtype=w.coord_features.dtype
|
125 |
+
).view(-1, 1)
|
126 |
+
coords = encode_coordinate(coords, w)
|
127 |
+
|
128 |
+
if sizes:
|
129 |
+
sizes = torch.tensor(
|
130 |
+
sizes, device=w.size_features.device, dtype=w.size_features.dtype
|
131 |
+
)
|
132 |
+
sizes = encode_size(sizes, w)
|
133 |
+
else:
|
134 |
+
sizes = None
|
135 |
+
|
136 |
+
return {"coords": coords, "sizes": sizes}
|
text.py
CHANGED
@@ -2,8 +2,9 @@ import torch
|
|
2 |
import torch.nn as nn
|
3 |
|
4 |
from torch.nn import functional as F
|
|
|
5 |
|
6 |
-
from .layers import layer_norm, mlp
|
7 |
from .rope import apply_rotary_emb, precompute_freqs_cis
|
8 |
from .config import TextConfig
|
9 |
|
@@ -21,25 +22,22 @@ def attn(
|
|
21 |
n_heads: int,
|
22 |
n_kv_heads: int,
|
23 |
position_ids: torch.Tensor,
|
|
|
24 |
):
|
25 |
bsz, q_len, d_model = x.shape
|
26 |
head_dim = d_model // n_heads
|
27 |
|
28 |
qkv_out = w.qkv(x) # shape: (bsz, q_len, (n_heads + 2*n_kv_heads)*head_dim)
|
|
|
|
|
29 |
q_dim = n_heads * head_dim
|
30 |
kv_dim = n_kv_heads * head_dim
|
|
|
|
|
31 |
|
32 |
-
q =
|
33 |
-
k = (
|
34 |
-
|
35 |
-
.view(bsz, q_len, n_kv_heads, head_dim)
|
36 |
-
.transpose(1, 2)
|
37 |
-
)
|
38 |
-
v = (
|
39 |
-
qkv_out[..., q_dim + kv_dim :]
|
40 |
-
.view(bsz, q_len, n_kv_heads, head_dim)
|
41 |
-
.transpose(1, 2)
|
42 |
-
)
|
43 |
|
44 |
q = apply_rotary_emb(q, freqs_cis, position_ids, n_heads)
|
45 |
k = apply_rotary_emb(k, freqs_cis, position_ids, n_kv_heads)
|
@@ -51,7 +49,14 @@ def attn(
|
|
51 |
q, k, v, attn_mask=attn_mask, enable_gqa=n_heads != n_kv_heads
|
52 |
)
|
53 |
out = out.transpose(1, 2).reshape(bsz, q_len, d_model)
|
54 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
55 |
return out
|
56 |
|
57 |
|
@@ -126,8 +131,17 @@ def text_decoder(
|
|
126 |
attn_mask: torch.Tensor,
|
127 |
position_ids: torch.Tensor,
|
128 |
config: TextConfig,
|
|
|
129 |
):
|
130 |
for i, block in enumerate(w.blocks):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
131 |
l_in = layer_norm(x, block.ln)
|
132 |
l_attn = attn(
|
133 |
l_in,
|
@@ -138,8 +152,9 @@ def text_decoder(
|
|
138 |
n_heads=config.n_heads,
|
139 |
n_kv_heads=config.n_kv_heads,
|
140 |
position_ids=position_ids,
|
|
|
141 |
)
|
142 |
-
l_mlp = mlp(l_in, block.mlp)
|
143 |
x = x + l_attn + l_mlp
|
144 |
|
145 |
return x
|
@@ -160,6 +175,7 @@ def _lm_head(hidden_BTC: torch.Tensor, w: nn.Module):
|
|
160 |
|
161 |
def build_text_model(config: TextConfig, dtype: torch.dtype) -> nn.Module:
|
162 |
qkv_dim = int(config.dim * (1 + 2 * config.n_kv_heads / config.n_heads))
|
|
|
163 |
|
164 |
text = nn.ModuleDict(
|
165 |
{
|
@@ -170,18 +186,18 @@ def build_text_model(config: TextConfig, dtype: torch.dtype) -> nn.Module:
|
|
170 |
"ln": nn.LayerNorm(config.dim, dtype=dtype),
|
171 |
"attn": nn.ModuleDict(
|
172 |
{
|
173 |
-
"qkv":
|
174 |
-
"proj":
|
175 |
config.dim, config.dim, dtype=dtype
|
176 |
),
|
177 |
}
|
178 |
),
|
179 |
"mlp": nn.ModuleDict(
|
180 |
{
|
181 |
-
"fc1":
|
182 |
config.dim, config.ff_dim, dtype=dtype
|
183 |
),
|
184 |
-
"fc2":
|
185 |
config.ff_dim, config.dim, dtype=dtype
|
186 |
),
|
187 |
}
|
|
|
2 |
import torch.nn as nn
|
3 |
|
4 |
from torch.nn import functional as F
|
5 |
+
from typing import Optional
|
6 |
|
7 |
+
from .layers import layer_norm, mlp, QuantizedLinear
|
8 |
from .rope import apply_rotary_emb, precompute_freqs_cis
|
9 |
from .config import TextConfig
|
10 |
|
|
|
22 |
n_heads: int,
|
23 |
n_kv_heads: int,
|
24 |
position_ids: torch.Tensor,
|
25 |
+
lora: Optional[dict],
|
26 |
):
|
27 |
bsz, q_len, d_model = x.shape
|
28 |
head_dim = d_model // n_heads
|
29 |
|
30 |
qkv_out = w.qkv(x) # shape: (bsz, q_len, (n_heads + 2*n_kv_heads)*head_dim)
|
31 |
+
if lora is not None:
|
32 |
+
qkv_out += F.linear(F.linear(x, lora["qkv"]["A"]), lora["qkv"]["B"])
|
33 |
q_dim = n_heads * head_dim
|
34 |
kv_dim = n_kv_heads * head_dim
|
35 |
+
q, k, v = qkv_out.split([q_dim, kv_dim, kv_dim], dim=-1)
|
36 |
+
del qkv_out
|
37 |
|
38 |
+
q = q.view(bsz, q_len, n_heads, head_dim).transpose(1, 2)
|
39 |
+
k = k.view(bsz, q_len, n_kv_heads, head_dim).transpose(1, 2)
|
40 |
+
v = v.view(bsz, q_len, n_kv_heads, head_dim).transpose(1, 2)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
|
42 |
q = apply_rotary_emb(q, freqs_cis, position_ids, n_heads)
|
43 |
k = apply_rotary_emb(k, freqs_cis, position_ids, n_kv_heads)
|
|
|
49 |
q, k, v, attn_mask=attn_mask, enable_gqa=n_heads != n_kv_heads
|
50 |
)
|
51 |
out = out.transpose(1, 2).reshape(bsz, q_len, d_model)
|
52 |
+
|
53 |
+
out0 = w.proj(out)
|
54 |
+
if lora is not None:
|
55 |
+
out1 = F.linear(F.linear(x, lora["proj"]["A"]), lora["proj"]["B"])
|
56 |
+
out = out0 + out1
|
57 |
+
else:
|
58 |
+
out = out0
|
59 |
+
|
60 |
return out
|
61 |
|
62 |
|
|
|
131 |
attn_mask: torch.Tensor,
|
132 |
position_ids: torch.Tensor,
|
133 |
config: TextConfig,
|
134 |
+
lora: Optional[dict],
|
135 |
):
|
136 |
for i, block in enumerate(w.blocks):
|
137 |
+
if lora is not None:
|
138 |
+
layer_lora = lora["text"]["blocks"][str(i)]
|
139 |
+
mlp_lora = layer_lora["mlp"]
|
140 |
+
attn_lora = layer_lora["attn"]
|
141 |
+
else:
|
142 |
+
mlp_lora = None
|
143 |
+
attn_lora = None
|
144 |
+
|
145 |
l_in = layer_norm(x, block.ln)
|
146 |
l_attn = attn(
|
147 |
l_in,
|
|
|
152 |
n_heads=config.n_heads,
|
153 |
n_kv_heads=config.n_kv_heads,
|
154 |
position_ids=position_ids,
|
155 |
+
lora=attn_lora,
|
156 |
)
|
157 |
+
l_mlp = mlp(l_in, block.mlp, lora=mlp_lora)
|
158 |
x = x + l_attn + l_mlp
|
159 |
|
160 |
return x
|
|
|
175 |
|
176 |
def build_text_model(config: TextConfig, dtype: torch.dtype) -> nn.Module:
|
177 |
qkv_dim = int(config.dim * (1 + 2 * config.n_kv_heads / config.n_heads))
|
178 |
+
linear_cls = QuantizedLinear if config.group_size is not None else nn.Linear
|
179 |
|
180 |
text = nn.ModuleDict(
|
181 |
{
|
|
|
186 |
"ln": nn.LayerNorm(config.dim, dtype=dtype),
|
187 |
"attn": nn.ModuleDict(
|
188 |
{
|
189 |
+
"qkv": linear_cls(config.dim, qkv_dim, dtype=dtype),
|
190 |
+
"proj": linear_cls(
|
191 |
config.dim, config.dim, dtype=dtype
|
192 |
),
|
193 |
}
|
194 |
),
|
195 |
"mlp": nn.ModuleDict(
|
196 |
{
|
197 |
+
"fc1": linear_cls(
|
198 |
config.dim, config.ff_dim, dtype=dtype
|
199 |
),
|
200 |
+
"fc2": linear_cls(
|
201 |
config.ff_dim, config.dim, dtype=dtype
|
202 |
),
|
203 |
}
|
vision.py
CHANGED
@@ -6,7 +6,7 @@ import numpy as np
|
|
6 |
from typing import Union, Tuple
|
7 |
from PIL import Image
|
8 |
|
9 |
-
from .layers import attn, layer_norm,
|
10 |
from .image_crops import overlap_crop_image
|
11 |
from .config import VisionConfig
|
12 |
|
@@ -33,7 +33,7 @@ def prepare_crops(
|
|
33 |
all_crops = np.transpose(all_crops, (0, 3, 1, 2))
|
34 |
all_crops = (
|
35 |
torch.from_numpy(all_crops)
|
36 |
-
.to(device=device, dtype=torch.
|
37 |
.div_(255.0)
|
38 |
.sub_(0.5)
|
39 |
.div_(0.5)
|
@@ -64,7 +64,7 @@ def create_patches(x, patch_size):
|
|
64 |
def vision_encoder(input_BCHW: torch.Tensor, w: nn.Module, config: VisionConfig):
|
65 |
x = create_patches(input_BCHW, config.enc_patch_size)
|
66 |
|
67 |
-
x =
|
68 |
x = x + w.pos_emb
|
69 |
for block in w.blocks:
|
70 |
x = x + attn(layer_norm(x, block.ln1), block.attn, n_heads=config.enc_n_heads)
|
|
|
6 |
from typing import Union, Tuple
|
7 |
from PIL import Image
|
8 |
|
9 |
+
from .layers import attn, layer_norm, mlp
|
10 |
from .image_crops import overlap_crop_image
|
11 |
from .config import VisionConfig
|
12 |
|
|
|
33 |
all_crops = np.transpose(all_crops, (0, 3, 1, 2))
|
34 |
all_crops = (
|
35 |
torch.from_numpy(all_crops)
|
36 |
+
.to(device=device, dtype=torch.bfloat16)
|
37 |
.div_(255.0)
|
38 |
.sub_(0.5)
|
39 |
.div_(0.5)
|
|
|
64 |
def vision_encoder(input_BCHW: torch.Tensor, w: nn.Module, config: VisionConfig):
|
65 |
x = create_patches(input_BCHW, config.enc_patch_size)
|
66 |
|
67 |
+
x = w.patch_emb(x)
|
68 |
x = x + w.pos_emb
|
69 |
for block in w.blocks:
|
70 |
x = x + attn(layer_norm(x, block.ln1), block.attn, n_heads=config.enc_n_heads)
|