Text Classification
Transformers
Safetensors
PEFT
branchone
feature-extraction
decision-model
multimodal
custom-code
custom_code
Instructions to use ndemirdemir/BranchOne with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use ndemirdemir/BranchOne with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-classification", model="ndemirdemir/BranchOne", trust_remote_code=True)# pip install -U transformers accelerate # Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("ndemirdemir/BranchOne", trust_remote_code=True, device_map="auto") - PEFT
How to use ndemirdemir/BranchOne with PEFT:
Task type is invalid.
- Notebooks
- Google Colab
- Kaggle
File size: 1,822 Bytes
f78d5a7 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 | """Shared candidate scorer; checkpoint parameter names are preserved."""
from torch import nn
TARGETS = [
"q_proj",
"k_proj",
"v_proj",
"o_proj",
"gate_proj",
"up_proj",
"down_proj",
"in_proj_qkv",
"in_proj_z",
"in_proj_a",
"in_proj_b",
"out_proj",
]
class CandidateHead(nn.Module):
def __init__(self, hidden=1024, width=256, comparison=False):
super().__init__()
self.width = width
self.comparison = comparison
# Common layers are initialized in the same order for both variants.
self.norm = nn.LayerNorm(hidden)
self.project = nn.Linear(hidden, width)
self.ff_norm = nn.LayerNorm(width)
self.ff = nn.Sequential(
nn.Linear(width, width * 2), nn.GELU(), nn.Linear(width * 2, width)
)
self.final_norm = nn.LayerNorm(width)
self.scorer = nn.Linear(width, 1)
if comparison:
self.set_norm = nn.LayerNorm(width)
self.attention = nn.MultiheadAttention(
width, 4, dropout=0.0, batch_first=True
)
# Start as the independent head; learn interactions without perturbing its initial logits.
nn.init.zeros_(self.attention.out_proj.weight)
nn.init.zeros_(self.attention.out_proj.bias)
def forward(self, h, valid, return_features=False):
u = self.project(self.norm(h.float()))
if self.comparison:
q = self.set_norm(u)
delta, _ = self.attention(
q, q, q, key_padding_mask=~valid, need_weights=False
)
u = u + delta
u = u + self.ff(self.ff_norm(u))
u = self.final_norm(u)
z = self.scorer(u).squeeze(-1).masked_fill(~valid, -1e9)
return (z, u) if return_features else z
|