tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

make e2e tests a bit faster by reducing test split size (#2522) [skip ci]

Browse Source 
* [ci] make e2e tests a bit faster by reducing test split size

* use 10% split of alpaca dataset to speed up dataset loading/tokenization

* reduce gas 4->2 for most e2e tests

* increase val set size for packing
This commit is contained in:
Wing Lian
2025-04-12 07:24:43 -07:00
committed by GitHub
parent 51267ded04
commit de8a625dd7
24 changed files with 54 additions and 44 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/e2e/integrations/test_cut_cross_entropy.py
View File

@@ -25,7 +25,7 @@ def min_cfg(temp_dir):
], ],
"cut_cross_entropy": True, "cut_cross_entropy": True,
"sequence_len": 1024, "sequence_len": 1024,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": { "special_tokens": {
"pad_token": "<|endoftext|>", "pad_token": "<|endoftext|>",
}, },
@@ -79,7 +79,7 @@ class TestCutCrossEntropyIntegration:
], ],
"cut_cross_entropy": True, "cut_cross_entropy": True,
"sequence_len": 1024, "sequence_len": 1024,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": { "special_tokens": {
"pad_token": "<|endoftext|>", "pad_token": "<|endoftext|>",
}, },

2
tests/e2e/multigpu/solo/test_flex.py
View File

@@ -55,7 +55,7 @@ class TestPackedFlex:
], ],
"num_epochs": 1, "num_epochs": 1,
"micro_batch_size": 2, "micro_batch_size": 2,
"gradient_accumulation_steps": 4, "gradient_accumulation_steps": 2,
"output_dir": temp_dir, "output_dir": temp_dir,
"learning_rate": 0.00001, "learning_rate": 0.00001,
"optimizer": "adamw_torch_fused", "optimizer": "adamw_torch_fused",

19
tests/e2e/multigpu/test_llama.py
View File

@@ -58,12 +58,13 @@ class TestMultiGPULlama:
{ {
"path": "tatsu-lab/alpaca", "path": "tatsu-lab/alpaca",
"type": "alpaca", "type": "alpaca",
"split": "train[:10%]",
}, },
], ],
"num_epochs": 1, "num_epochs": 1,
"max_steps": 2, "max_steps": 2,
"micro_batch_size": 1, "micro_batch_size": 1,
"gradient_accumulation_steps": 4, "gradient_accumulation_steps": 2,
# "gradient_checkpointing": True, # "gradient_checkpointing": True,
"output_dir": temp_dir, "output_dir": temp_dir,
"learning_rate": 0.00001, "learning_rate": 0.00001,
@@ -201,7 +202,7 @@ class TestMultiGPULlama:
"num_epochs": 1, "num_epochs": 1,
"max_steps": 2, "max_steps": 2,
"micro_batch_size": 2, "micro_batch_size": 2,
"gradient_accumulation_steps": 4, "gradient_accumulation_steps": 2,
# "gradient_checkpointing": True, # "gradient_checkpointing": True,
"output_dir": temp_dir, "output_dir": temp_dir,
"warmup_steps": 0, "warmup_steps": 0,
@@ -279,7 +280,7 @@ class TestMultiGPULlama:
"num_epochs": 1, "num_epochs": 1,
"max_steps": 2, "max_steps": 2,
"micro_batch_size": 2, "micro_batch_size": 2,
"gradient_accumulation_steps": 4, "gradient_accumulation_steps": 2,
# "gradient_checkpointing": True, # "gradient_checkpointing": True,
"output_dir": temp_dir, "output_dir": temp_dir,
"warmup_steps": 0, "warmup_steps": 0,
@@ -335,6 +336,7 @@ class TestMultiGPULlama:
{ {
"path": "tatsu-lab/alpaca", "path": "tatsu-lab/alpaca",
"type": "alpaca", "type": "alpaca",
"split": "train[:10%]",
}, },
], ],
"num_epochs": 1, "num_epochs": 1,
@@ -398,7 +400,7 @@ class TestMultiGPULlama:
"sample_packing": True, "sample_packing": True,
"pad_to_sequence_len": True, "pad_to_sequence_len": True,
"sequence_len": 1024, "sequence_len": 1024,
"val_set_size": 0.01, "val_set_size": 0.05,
"special_tokens": { "special_tokens": {
"pad_token": "<|endoftext|>", "pad_token": "<|endoftext|>",
}, },
@@ -406,6 +408,7 @@ class TestMultiGPULlama:
{ {
"path": "tatsu-lab/alpaca", "path": "tatsu-lab/alpaca",
"type": "alpaca", "type": "alpaca",
"split": "train[:10%]",
}, },
], ],
"num_epochs": 1, "num_epochs": 1,
@@ -484,6 +487,7 @@ class TestMultiGPULlama:
{ {
"path": "tatsu-lab/alpaca", "path": "tatsu-lab/alpaca",
"type": "alpaca", "type": "alpaca",
"split": "train[:10%]",
}, },
], ],
"num_epochs": 1, "num_epochs": 1,
@@ -565,7 +569,7 @@ class TestMultiGPULlama:
{ {
"path": "tatsu-lab/alpaca", "path": "tatsu-lab/alpaca",
"type": "alpaca", "type": "alpaca",
"split": "train[:25%]", "split": "train[:10%]",
}, },
], ],
"num_epochs": 1, "num_epochs": 1,
@@ -660,7 +664,7 @@ class TestMultiGPULlama:
"sample_packing": True, "sample_packing": True,
"pad_to_sequence_len": True, "pad_to_sequence_len": True,
"sequence_len": 1024, "sequence_len": 1024,
"val_set_size": 0.01, "val_set_size": 0.05,
"special_tokens": { "special_tokens": {
"pad_token": "<|endoftext|>", "pad_token": "<|endoftext|>",
}, },
@@ -668,6 +672,7 @@ class TestMultiGPULlama:
{ {
"path": "tatsu-lab/alpaca", "path": "tatsu-lab/alpaca",
"type": "alpaca", "type": "alpaca",
"split": "train[:10%]",
}, },
], ],
"num_epochs": 1, "num_epochs": 1,
@@ -741,6 +746,7 @@ class TestMultiGPULlama:
{ {
"path": "tatsu-lab/alpaca", "path": "tatsu-lab/alpaca",
"type": "alpaca", "type": "alpaca",
"split": "train[:10%]",
}, },
], ],
"num_epochs": 1, "num_epochs": 1,
@@ -814,6 +820,7 @@ class TestMultiGPULlama:
{ {
"path": "tatsu-lab/alpaca", "path": "tatsu-lab/alpaca",
"type": "alpaca", "type": "alpaca",
"split": "train[:10%]",
}, },
], ],
"num_epochs": 1, "num_epochs": 1,

2
tests/e2e/multigpu/test_ray.py
View File

@@ -45,6 +45,7 @@ class TestMultiGPURay:
{ {
"path": "tatsu-lab/alpaca", "path": "tatsu-lab/alpaca",
"type": "alpaca", "type": "alpaca",
"split": "train[:10%]",
}, },
], ],
"num_epochs": 1, "num_epochs": 1,
@@ -103,6 +104,7 @@ class TestMultiGPURay:
{ {
"path": "tatsu-lab/alpaca", "path": "tatsu-lab/alpaca",
"type": "alpaca", "type": "alpaca",
"split": "train[:10%]",
}, },
], ],
"num_epochs": 1, "num_epochs": 1,

1
tests/e2e/multigpu/test_sp.py
View File

@@ -40,6 +40,7 @@ class TestSequenceParallelism:
{ {
"path": "tatsu-lab/alpaca", "path": "tatsu-lab/alpaca",
"type": "alpaca", "type": "alpaca",
"split": "train[:10%]",
}, },
], ],
"num_epochs": 1, "num_epochs": 1,

4
tests/e2e/patched/test_llama_s2_attention.py
View File

@@ -43,7 +43,7 @@ class TestLlamaShiftedSparseAttention(unittest.TestCase):
"lora_alpha": 16, "lora_alpha": 16,
"lora_dropout": 0.05, "lora_dropout": 0.05,
"lora_target_linear": True, "lora_target_linear": True,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": {}, "special_tokens": {},
"datasets": [ "datasets": [
{ {
@@ -83,7 +83,7 @@ class TestLlamaShiftedSparseAttention(unittest.TestCase):
"sample_packing": False, "sample_packing": False,
"flash_attention": True, "flash_attention": True,
"s2_attention": True, "s2_attention": True,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": {}, "special_tokens": {},
"datasets": [ "datasets": [
{ {

4
tests/e2e/patched/test_model_patches.py
View File

@@ -27,7 +27,7 @@ class TestModelPatches(unittest.TestCase):
"flash_attention": True, "flash_attention": True,
"sample_packing": True, "sample_packing": True,
"sequence_len": 2048, "sequence_len": 2048,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": {}, "special_tokens": {},
"datasets": [ "datasets": [
{ {
@@ -59,7 +59,7 @@ class TestModelPatches(unittest.TestCase):
"flash_attention": True, "flash_attention": True,
"sample_packing": True, "sample_packing": True,
"sequence_len": 2048, "sequence_len": 2048,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": {}, "special_tokens": {},
"datasets": [ "datasets": [
{ {

2
tests/e2e/patched/test_phi_multipack.py
View File

@@ -88,7 +88,7 @@ class TestPhiMultipack(unittest.TestCase):
"lora_alpha": 32, "lora_alpha": 32,
"lora_dropout": 0.05, "lora_dropout": 0.05,
"lora_target_linear": True, "lora_target_linear": True,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": { "special_tokens": {
"pad_token": "<|endoftext|>", "pad_token": "<|endoftext|>",
}, },

2
tests/e2e/solo/test_flex.py
View File

@@ -47,7 +47,7 @@ class TestPackedFlex(unittest.TestCase):
], ],
"num_epochs": 1, "num_epochs": 1,
"micro_batch_size": 2, "micro_batch_size": 2,
"gradient_accumulation_steps": 4, "gradient_accumulation_steps": 2,
"output_dir": temp_dir, "output_dir": temp_dir,
"learning_rate": 0.00001, "learning_rate": 0.00001,
"optimizer": "adamw_torch_fused", "optimizer": "adamw_torch_fused",

4
tests/e2e/test_deepseekv3.py
View File

@@ -65,7 +65,7 @@ class TestDeepseekV3:
"chat_template": "deepseek_v3", "chat_template": "deepseek_v3",
"num_epochs": 1, "num_epochs": 1,
"micro_batch_size": 1, "micro_batch_size": 1,
"gradient_accumulation_steps": 4, "gradient_accumulation_steps": 2,
"output_dir": temp_dir, "output_dir": temp_dir,
"learning_rate": 0.00001, "learning_rate": 0.00001,
"optimizer": "adamw_bnb_8bit", "optimizer": "adamw_bnb_8bit",
@@ -115,7 +115,7 @@ class TestDeepseekV3:
}, },
"num_epochs": 1, "num_epochs": 1,
"micro_batch_size": 1, "micro_batch_size": 1,
"gradient_accumulation_steps": 4, "gradient_accumulation_steps": 2,
"output_dir": temp_dir, "output_dir": temp_dir,
"learning_rate": 0.00001, "learning_rate": 0.00001,
"optimizer": "adamw_bnb_8bit", "optimizer": "adamw_bnb_8bit",

6
tests/e2e/test_falcon.py
View File

@@ -41,7 +41,7 @@ class TestFalcon(unittest.TestCase):
"word_embeddings", "word_embeddings",
"lm_head", "lm_head",
], ],
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": { "special_tokens": {
"bos_token": "<|endoftext|>", "bos_token": "<|endoftext|>",
"pad_token": "<|endoftext|>", "pad_token": "<|endoftext|>",
@@ -92,7 +92,7 @@ class TestFalcon(unittest.TestCase):
"word_embeddings", "word_embeddings",
"lm_head", "lm_head",
], ],
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": { "special_tokens": {
"bos_token": "<|endoftext|>", "bos_token": "<|endoftext|>",
"pad_token": "<|endoftext|>", "pad_token": "<|endoftext|>",
@@ -137,7 +137,7 @@ class TestFalcon(unittest.TestCase):
"base_model": "illuin/tiny-random-FalconForCausalLM", "base_model": "illuin/tiny-random-FalconForCausalLM",
"flash_attention": True, "flash_attention": True,
"sequence_len": 1024, "sequence_len": 1024,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": { "special_tokens": {
"bos_token": "<|endoftext|>", "bos_token": "<|endoftext|>",
"pad_token": "<|endoftext|>", "pad_token": "<|endoftext|>",

4
tests/e2e/test_gemma2.py
View File

@@ -62,7 +62,7 @@ class TestGemma2:
"chat_template": "gemma", # gemma2's template is same as gemma "chat_template": "gemma", # gemma2's template is same as gemma
"num_epochs": 1, "num_epochs": 1,
"micro_batch_size": 1, "micro_batch_size": 1,
"gradient_accumulation_steps": 4, "gradient_accumulation_steps": 2,
"output_dir": temp_dir, "output_dir": temp_dir,
"learning_rate": 0.00001, "learning_rate": 0.00001,
"optimizer": "adamw_bnb_8bit", "optimizer": "adamw_bnb_8bit",
@@ -114,7 +114,7 @@ class TestGemma2:
}, },
"num_epochs": 1, "num_epochs": 1,
"micro_batch_size": 1, "micro_batch_size": 1,
"gradient_accumulation_steps": 4, "gradient_accumulation_steps": 2,
"output_dir": temp_dir, "output_dir": temp_dir,
"learning_rate": 0.00001, "learning_rate": 0.00001,
"optimizer": "adamw_bnb_8bit", "optimizer": "adamw_bnb_8bit",

4
tests/e2e/test_gemma3_text.py
View File

@@ -61,7 +61,7 @@ class TestGemma3Text:
"chat_template": "gemma3", "chat_template": "gemma3",
"num_epochs": 1, "num_epochs": 1,
"micro_batch_size": 1, "micro_batch_size": 1,
"gradient_accumulation_steps": 4, "gradient_accumulation_steps": 2,
"output_dir": temp_dir, "output_dir": temp_dir,
"learning_rate": 0.00001, "learning_rate": 0.00001,
"optimizer": "adamw_bnb_8bit", "optimizer": "adamw_bnb_8bit",
@@ -112,7 +112,7 @@ class TestGemma3Text:
}, },
"num_epochs": 1, "num_epochs": 1,
"micro_batch_size": 1, "micro_batch_size": 1,
"gradient_accumulation_steps": 4, "gradient_accumulation_steps": 2,
"output_dir": temp_dir, "output_dir": temp_dir,
"learning_rate": 0.00001, "learning_rate": 0.00001,
"optimizer": "adamw_bnb_8bit", "optimizer": "adamw_bnb_8bit",

2
tests/e2e/test_llama.py
View File

@@ -30,7 +30,7 @@ class TestLlama:
"tokenizer_type": "LlamaTokenizer", "tokenizer_type": "LlamaTokenizer",
"trust_remote_code": True, "trust_remote_code": True,
"sequence_len": 512, "sequence_len": 512,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": { "special_tokens": {
"unk_token": "<unk>", "unk_token": "<unk>",
"bos_token": "<s>", "bos_token": "<s>",

4
tests/e2e/test_llama_vision.py
View File

@@ -52,7 +52,7 @@ class TestLlamaVision(unittest.TestCase):
], ],
"num_epochs": 1, "num_epochs": 1,
"micro_batch_size": 1, "micro_batch_size": 1,
"gradient_accumulation_steps": 4, "gradient_accumulation_steps": 2,
"output_dir": temp_dir, "output_dir": temp_dir,
"learning_rate": 0.00001, "learning_rate": 0.00001,
"optimizer": "adamw_bnb_8bit", "optimizer": "adamw_bnb_8bit",
@@ -99,7 +99,7 @@ class TestLlamaVision(unittest.TestCase):
], ],
"num_epochs": 1, "num_epochs": 1,
"micro_batch_size": 1, "micro_batch_size": 1,
"gradient_accumulation_steps": 4, "gradient_accumulation_steps": 2,
"output_dir": temp_dir, "output_dir": temp_dir,
"learning_rate": 0.00001, "learning_rate": 0.00001,
"optimizer": "adamw_bnb_8bit", "optimizer": "adamw_bnb_8bit",

2
tests/e2e/test_load_model.py
View File

@@ -36,7 +36,7 @@ class TestLoadModelUtils:
"lora_alpha": 16, "lora_alpha": 16,
"lora_dropout": 0.05, "lora_dropout": 0.05,
"lora_target_linear": True, "lora_target_linear": True,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": { "special_tokens": {
"unk_token": "<unk>", "unk_token": "<unk>",
"bos_token": "<s>", "bos_token": "<s>",

2
tests/e2e/test_lora_llama.py
View File

@@ -37,7 +37,7 @@ class TestLoraLlama(unittest.TestCase):
"lora_alpha": 16, "lora_alpha": 16,
"lora_dropout": 0.05, "lora_dropout": 0.05,
"lora_target_linear": True, "lora_target_linear": True,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": { "special_tokens": {
"unk_token": "<unk>", "unk_token": "<unk>",
"bos_token": "<s>", "bos_token": "<s>",

4
tests/e2e/test_mistral.py
View File

@@ -39,7 +39,7 @@ class TestMistral(unittest.TestCase):
"lora_alpha": 64, "lora_alpha": 64,
"lora_dropout": 0.05, "lora_dropout": 0.05,
"lora_target_linear": True, "lora_target_linear": True,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": { "special_tokens": {
"unk_token": "<unk>", "unk_token": "<unk>",
"bos_token": "<s>", "bos_token": "<s>",
@@ -80,7 +80,7 @@ class TestMistral(unittest.TestCase):
"base_model": "openaccess-ai-collective/tiny-mistral", "base_model": "openaccess-ai-collective/tiny-mistral",
"flash_attention": True, "flash_attention": True,
"sequence_len": 1024, "sequence_len": 1024,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": { "special_tokens": {
"unk_token": "<unk>", "unk_token": "<unk>",
"bos_token": "<s>", "bos_token": "<s>",

10
tests/e2e/test_mixtral.py
View File

@@ -49,7 +49,7 @@ class TestMixtral(unittest.TestCase):
"q_proj", "q_proj",
"w2", "w2",
], ],
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": {}, "special_tokens": {},
"datasets": [ "datasets": [
{ {
@@ -105,7 +105,7 @@ class TestMixtral(unittest.TestCase):
"q_proj", "q_proj",
"w2", "w2",
], ],
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": {}, "special_tokens": {},
"datasets": [ "datasets": [
{ {
@@ -160,7 +160,7 @@ class TestMixtral(unittest.TestCase):
"q_proj", "q_proj",
"w2", "w2",
], ],
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": {}, "special_tokens": {},
"datasets": [ "datasets": [
{ {
@@ -219,7 +219,7 @@ class TestMixtral(unittest.TestCase):
"q_proj", "q_proj",
"w2", "w2",
], ],
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": {}, "special_tokens": {},
"datasets": [ "datasets": [
{ {
@@ -265,7 +265,7 @@ class TestMixtral(unittest.TestCase):
"tokenizer_config": "LoneStriker/Mixtral-8x7B-v0.1-HF", "tokenizer_config": "LoneStriker/Mixtral-8x7B-v0.1-HF",
"flash_attention": True, "flash_attention": True,
"sequence_len": 1024, "sequence_len": 1024,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": {}, "special_tokens": {},
"datasets": [ "datasets": [
{ {

6
tests/e2e/test_optimizers.py
View File

@@ -37,7 +37,7 @@ class TestCustomOptimizers(unittest.TestCase):
"lora_alpha": 16, "lora_alpha": 16,
"lora_dropout": 0.05, "lora_dropout": 0.05,
"lora_target_linear": True, "lora_target_linear": True,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": { "special_tokens": {
"unk_token": "<unk>", "unk_token": "<unk>",
"bos_token": "<s>", "bos_token": "<s>",
@@ -84,7 +84,7 @@ class TestCustomOptimizers(unittest.TestCase):
"lora_alpha": 16, "lora_alpha": 16,
"lora_dropout": 0.05, "lora_dropout": 0.05,
"lora_target_linear": True, "lora_target_linear": True,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": { "special_tokens": {
"unk_token": "<unk>", "unk_token": "<unk>",
"bos_token": "<s>", "bos_token": "<s>",
@@ -131,7 +131,7 @@ class TestCustomOptimizers(unittest.TestCase):
"lora_alpha": 16, "lora_alpha": 16,
"lora_dropout": 0.05, "lora_dropout": 0.05,
"lora_target_linear": True, "lora_target_linear": True,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": { "special_tokens": {
"unk_token": "<unk>", "unk_token": "<unk>",
"bos_token": "<s>", "bos_token": "<s>",

2
tests/e2e/test_packing_loss.py
View File

@@ -46,7 +46,7 @@ class TestPackedLlama(unittest.TestCase):
], ],
"num_epochs": 1, "num_epochs": 1,
"micro_batch_size": 2, "micro_batch_size": 2,
"gradient_accumulation_steps": 4, "gradient_accumulation_steps": 2,
"output_dir": temp_dir, "output_dir": temp_dir,
"learning_rate": 0.00001, "learning_rate": 0.00001,
"optimizer": "adamw_torch_fused", "optimizer": "adamw_torch_fused",

4
tests/e2e/test_phi.py
View File

@@ -35,7 +35,7 @@ class TestPhi(unittest.TestCase):
"sample_packing": False, "sample_packing": False,
"load_in_8bit": False, "load_in_8bit": False,
"adapter": None, "adapter": None,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": { "special_tokens": {
"pad_token": "<|endoftext|>", "pad_token": "<|endoftext|>",
}, },
@@ -85,7 +85,7 @@ class TestPhi(unittest.TestCase):
"lora_alpha": 32, "lora_alpha": 32,
"lora_dropout": 0.05, "lora_dropout": 0.05,
"lora_target_linear": True, "lora_target_linear": True,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": { "special_tokens": {
"pad_token": "<|endoftext|>", "pad_token": "<|endoftext|>",
}, },

2
tests/e2e/test_schedulers.py
View File

@@ -37,7 +37,7 @@ class TestCustomSchedulers(unittest.TestCase):
"lora_alpha": 16, "lora_alpha": 16,
"lora_dropout": 0.05, "lora_dropout": 0.05,
"lora_target_linear": True, "lora_target_linear": True,
"val_set_size": 0.1, "val_set_size": 0.02,
"special_tokens": { "special_tokens": {
"unk_token": "<unk>", "unk_token": "<unk>",
"bos_token": "<s>", "bos_token": "<s>",

2
tests/test_exact_deduplication.py
View File

@@ -313,7 +313,7 @@ class TestDeduplicateNonRL(unittest.TestCase):
}, },
], ],
"val_set_size": 0.0, "val_set_size": 0.0,
"gradient_accumulation_steps": 4, "gradient_accumulation_steps": 2,
"batch_size": 10, "batch_size": 10,
"micro_batch_size": 10, "micro_batch_size": 10,
"num_epochs": 1, "num_epochs": 1,