Andrewhinh's workspace
Runs
3
State
Notes
User
Tags
Created
Runtime
Sweep
_attn_implementation_autoset
_name_or_path
accelerator_config.even_batches
accelerator_config.non_blocking
accelerator_config.split_batches
accelerator_config.use_seedable_sampler
adafactor
adam_beta1
adam_beta2
adam_epsilon
add_cross_attention
amp
amp_dtype
amp_impl
architectures
attention_dropout
aug_repeats
aug_splits
auto_find_batch_size
average_tokens_across_devices
batch_eval_metrics
batch_size
bce_loss
bce_sum
bf16
bf16_full_eval
bos_token_id
channels_last
checkpoint_hist
chunk_size_feed_forward
clip_grad
clip_mode
color_jitter
cooldown_epochs
cutmix
data_args.buffer_size
data_args.cutoff_len
data_args.dataset
data_args.dataset_dir
data_args.ignore_pad_token_for_loss
data_args.image_dir
data_args.mask_history
data_args.max_samples
data_args.media_dir
Finished
-
andrewhinh
1h 6m 32s
-
true
andrewhinh/formless-qwen2.5-vl-7b-instruct-full-sft
true
false
false
true
false
0.9
0.999
1.0000e-8
false
-
-
-
["Qwen2_5_VLForConditionalGeneration"]
0
-
-
false
false
false
-
-
-
true
false
151643
-
-
0
-
-
-
-
-
16384
32768
["dpo"]
data
true
-
false
10000
data
Finished
-
andrewhinh
14m 3s
-
true
Qwen/Qwen2.5-VL-7B-Instruct
true
false
false
true
false
0.9
0.999
1.0000e-8
false
-
-
-
["Qwen2_5_VLForConditionalGeneration"]
0
-
-
false
false
false
-
-
-
true
false
151643
-
-
0
-
-
-
-
-
16384
32768
["sft"]
data
true
-
false
1000
data
Finished
-
andrewhinh
2h 1m 15s
-
-
-
-
-
-
-
-
-
-
-
-
true
bfloat16
native
-
-
0
0
-
-
-
128
false
false
-
-
-
false
1
-
1
norm
0.4
0
0
-
-
-
-
-
-
-
-
-
1-3
of 3