Code
Hub
Workspaces
Following
Trending
Connect
MCP
copy
Create free account
hub
/
github.com/andrew-cr/discrete_flow_models
/ functions
Functions
31 in github.com/andrew-cr/discrete_flow_models
⨍
Functions
31
◇
Types & classes
6
↓ 6 callers
Method
_run_net
(self, idx, time, x1=None, attn_mask=None)
flow_model.py:216
↓ 6 callers
Function
corrupt_data
(data, times)
train.py:303
↓ 5 callers
Function
get_batch
(split, times=None)
train.py:326
↓ 4 callers
Method
__init__
(self, config)
flow_model.py:93
↓ 4 callers
Function
decode
(l)
train.py:185
↓ 2 callers
Function
calc_loss
(X, Y, times, target_mask, infill_probs, num_ones_in_mask)
train.py:355
↓ 2 callers
Function
encode
(s)
data/text8/prepare.py:25
↓ 2 callers
Method
get_num_params
Return the number of parameters in the model. For non-embedding count (default), the position embeddings get subtracted. The
flow_model.py:196
↓ 2 callers
Function
save_checkpoint
(file_path)
train.py:486
↓ 2 callers
Function
transformer_timestep_embedding
(timesteps, embedding_dim, max_positions=10000)
flow_model.py:123
↓ 1 callers
Method
configure_optimizers
(self, weight_decay, learning_rate, betas, device_type)
flow_model.py:295
↓ 1 callers
Function
decode
(l)
sample.py:108
↓ 1 callers
Function
estimate_loss
()
train.py:372
↓ 1 callers
Method
estimate_mfu
estimate model flops utilization (MFU) in units of A100 bfloat16 peak FLOPS
flow_model.py:321
↓ 1 callers
Method
forward
(self, x)
flow_model.py:100
↓ 1 callers
Function
get_lr
(it)
train.py:394
↓ 1 callers
Function
load_model
(ckpt_path)
sample.py:116
Method
__init__
(self, ndim, bias)
flow_model.py:21
Method
__init__
(self, config)
flow_model.py:31
Method
__init__
(self, config)
flow_model.py:109
Method
__init__
config.vocab_size should include a mask token
flow_model.py:159
Method
_init_weights
(self, module)
flow_model.py:208
Function
decode
(l)
data/text8/prepare.py:27
Function
decode
(l)
eval/sample_eval.py:53
Function
encode
(s)
train.py:183
Function
encode
(s)
sample.py:106
Function
encode
(s)
eval/sample_eval.py:51
Method
forward
(self, input)
flow_model.py:26
Method
forward
(self, x, attn_mask=None)
flow_model.py:59
Method
forward
(self, x, attn_mask=None)
flow_model.py:116
Method
forward
idx is the corrupted tokens (b, t) time is the time in the corruption process (b,) targets is the clean data (b,
flow_model.py:248