Upload 80 files

save_lora_instruct_r8_3e4_s11_1/README.md

@@ -0,0 +1,202 @@
+---
+base_model: meta-llama/Llama-2-7b-hf
+library_name: peft
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.9.0

save_lora_instruct_r8_3e4_s11_1/adapter_config.json

@@ -0,0 +1,35 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "meta-llama/Llama-2-7b-hf",
+  "bias": "none",
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 16,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "mora_type": 1,
+  "peft_type": "LORA",
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "k_proj",
+    "v_proj",
+    "o_proj",
+    "gate_proj",
+    "down_proj",
+    "up_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_mora": false,
+  "use_rslora": false
+}

save_lora_instruct_r8_3e4_s11_1/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7da849593de343a0a6b1a5a0ad86c33518dc2b8d55d65fc9bd0e2d97574f95b0
+size 564342240

save_lora_instruct_r8_3e4_s11_1/checkpoint-2000/README.md

@@ -0,0 +1,202 @@
+---
+base_model: meta-llama/Llama-2-7b-hf
+library_name: peft
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.9.0

save_lora_instruct_r8_3e4_s11_1/checkpoint-2000/adapter_config.json

@@ -0,0 +1,35 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "meta-llama/Llama-2-7b-hf",
+  "bias": "none",
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 16,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "mora_type": 1,
+  "peft_type": "LORA",
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "k_proj",
+    "v_proj",
+    "o_proj",
+    "gate_proj",
+    "down_proj",
+    "up_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_mora": false,
+  "use_rslora": false
+}

save_lora_instruct_r8_3e4_s11_1/checkpoint-2000/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:88ef2211549a31cd7ff822db59b478bdaf2fa760c661c681cae33eef782f3fa1
+size 564342240

save_lora_instruct_r8_3e4_s11_1/checkpoint-2000/global_step2000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0aa301c54b9777df8a052c9a98b04f1aebb855233981d1589840abd9989ca50
+size 239914224

save_lora_instruct_r8_3e4_s11_1/checkpoint-2000/global_step2000/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b6af73b6acfa02d1aa0227c384704bc84e44823491227790e2a009721a52553
+size 40303404

save_lora_instruct_r8_3e4_s11_1/checkpoint-2000/latest

@@ -0,0 +1 @@
+global_step2000

save_lora_instruct_r8_3e4_s11_1/checkpoint-2000/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:120654c40ab07680b6e25c6f415f868d33c21d42ff816d413cf88a5c5793ca4b
+size 14244

save_lora_instruct_r8_3e4_s11_1/checkpoint-2000/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb0ea8bedbf8a7566772545687fcd7fa657898d5dc376764026636ff8b08022d
+size 1064

save_lora_instruct_r8_3e4_s11_1/checkpoint-2000/trainer_state.json

@@ -0,0 +1,2421 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.7983284997037453,
+  "eval_steps": 500,
+  "global_step": 2000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.999999999999999e-06,
+      "loss": 1.38,
+      "step": 5
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 1.9999999999999998e-05,
+      "loss": 1.3936,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 2.9999999999999997e-05,
+      "loss": 1.3006,
+      "step": 15
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.9999999999999996e-05,
+      "loss": 1.2594,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 4.9999999999999996e-05,
+      "loss": 1.1177,
+      "step": 25
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 5.9999999999999995e-05,
+      "loss": 1.0481,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 7e-05,
+      "loss": 1.0601,
+      "step": 35
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 7.999999999999999e-05,
+      "loss": 1.0473,
+      "step": 40
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 8.999999999999999e-05,
+      "loss": 0.9523,
+      "step": 45
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.999999999999999e-05,
+      "loss": 0.9451,
+      "step": 50
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00010999999999999998,
+      "loss": 0.9687,
+      "step": 55
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00011999999999999999,
+      "loss": 0.9054,
+      "step": 60
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00013,
+      "loss": 0.9518,
+      "step": 65
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00014,
+      "loss": 0.8763,
+      "step": 70
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00015,
+      "loss": 0.8657,
+      "step": 75
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00015999999999999999,
+      "loss": 0.9145,
+      "step": 80
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00016999999999999999,
+      "loss": 0.8296,
+      "step": 85
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00017999999999999998,
+      "loss": 0.8709,
+      "step": 90
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00018999999999999998,
+      "loss": 0.8405,
+      "step": 95
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00019999999999999998,
+      "loss": 0.8604,
+      "step": 100
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00020999999999999998,
+      "loss": 0.8607,
+      "step": 105
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00021999999999999995,
+      "loss": 0.8376,
+      "step": 110
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00023,
+      "loss": 0.8554,
+      "step": 115
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00023999999999999998,
+      "loss": 0.8456,
+      "step": 120
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00025,
+      "loss": 0.8443,
+      "step": 125
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00026,
+      "loss": 0.8327,
+      "step": 130
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00027,
+      "loss": 0.8096,
+      "step": 135
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00028,
+      "loss": 0.7972,
+      "step": 140
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00029,
+      "loss": 0.8306,
+      "step": 145
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0003,
+      "loss": 0.8409,
+      "step": 150
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0002996913580246913,
+      "loss": 0.8094,
+      "step": 155
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00029938271604938266,
+      "loss": 0.8146,
+      "step": 160
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00029907407407407405,
+      "loss": 0.8042,
+      "step": 165
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0002987654320987654,
+      "loss": 0.7784,
+      "step": 170
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0002984567901234568,
+      "loss": 0.828,
+      "step": 175
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00029814814814814813,
+      "loss": 0.8473,
+      "step": 180
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0002978395061728395,
+      "loss": 0.8354,
+      "step": 185
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00029753086419753087,
+      "loss": 0.8246,
+      "step": 190
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00029722222222222216,
+      "loss": 0.8249,
+      "step": 195
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00029691358024691355,
+      "loss": 0.8222,
+      "step": 200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0002966049382716049,
+      "loss": 0.8369,
+      "step": 205
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00029629629629629624,
+      "loss": 0.7954,
+      "step": 210
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00029598765432098763,
+      "loss": 0.8051,
+      "step": 215
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000295679012345679,
+      "loss": 0.8081,
+      "step": 220
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00029537037037037037,
+      "loss": 0.7864,
+      "step": 225
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0002950617283950617,
+      "loss": 0.7974,
+      "step": 230
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00029475308641975306,
+      "loss": 0.8306,
+      "step": 235
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00029444444444444445,
+      "loss": 0.7792,
+      "step": 240
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00029413580246913574,
+      "loss": 0.8108,
+      "step": 245
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00029382716049382714,
+      "loss": 0.8167,
+      "step": 250
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0002935185185185185,
+      "loss": 0.8358,
+      "step": 255
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0002932098765432099,
+      "loss": 0.7948,
+      "step": 260
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0002929012345679012,
+      "loss": 0.8078,
+      "step": 265
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00029259259259259256,
+      "loss": 0.7886,
+      "step": 270
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00029228395061728395,
+      "loss": 0.7903,
+      "step": 275
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0002919753086419753,
+      "loss": 0.7892,
+      "step": 280
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00029166666666666664,
+      "loss": 0.7717,
+      "step": 285
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.000291358024691358,
+      "loss": 0.7774,
+      "step": 290
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0002910493827160494,
+      "loss": 0.8053,
+      "step": 295
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0002907407407407407,
+      "loss": 0.794,
+      "step": 300
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00029043209876543206,
+      "loss": 0.7856,
+      "step": 305
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00029012345679012345,
+      "loss": 0.7713,
+      "step": 310
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0002898148148148148,
+      "loss": 0.7805,
+      "step": 315
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00028950617283950614,
+      "loss": 0.7897,
+      "step": 320
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00028919753086419753,
+      "loss": 0.7931,
+      "step": 325
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0002888888888888888,
+      "loss": 0.7739,
+      "step": 330
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0002885802469135802,
+      "loss": 0.7749,
+      "step": 335
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00028827160493827156,
+      "loss": 0.7932,
+      "step": 340
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00028796296296296296,
+      "loss": 0.7906,
+      "step": 345
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0002876543209876543,
+      "loss": 0.8094,
+      "step": 350
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00028734567901234564,
+      "loss": 0.7932,
+      "step": 355
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00028703703703703703,
+      "loss": 0.8027,
+      "step": 360
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0002867283950617284,
+      "loss": 0.8297,
+      "step": 365
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0002864197530864197,
+      "loss": 0.8228,
+      "step": 370
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0002861111111111111,
+      "loss": 0.7997,
+      "step": 375
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00028580246913580246,
+      "loss": 0.8124,
+      "step": 380
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0002854938271604938,
+      "loss": 0.8152,
+      "step": 385
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00028518518518518514,
+      "loss": 0.7833,
+      "step": 390
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00028487654320987654,
+      "loss": 0.7884,
+      "step": 395
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0002845679012345679,
+      "loss": 0.753,
+      "step": 400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0002842592592592592,
+      "loss": 0.8055,
+      "step": 405
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0002839506172839506,
+      "loss": 0.8014,
+      "step": 410
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00028364197530864196,
+      "loss": 0.782,
+      "step": 415
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0002833333333333333,
+      "loss": 0.7733,
+      "step": 420
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0002830246913580247,
+      "loss": 0.7777,
+      "step": 425
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00028271604938271604,
+      "loss": 0.752,
+      "step": 430
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0002824074074074074,
+      "loss": 0.7963,
+      "step": 435
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0002820987654320987,
+      "loss": 0.7655,
+      "step": 440
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0002817901234567901,
+      "loss": 0.8,
+      "step": 445
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00028148148148148146,
+      "loss": 0.7465,
+      "step": 450
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0002811728395061728,
+      "loss": 0.7432,
+      "step": 455
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0002808641975308642,
+      "loss": 0.758,
+      "step": 460
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00028055555555555554,
+      "loss": 0.7757,
+      "step": 465
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0002802469135802469,
+      "loss": 0.7816,
+      "step": 470
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0002799382716049382,
+      "loss": 0.7744,
+      "step": 475
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0002796296296296296,
+      "loss": 0.7807,
+      "step": 480
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00027932098765432096,
+      "loss": 0.774,
+      "step": 485
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0002790123456790123,
+      "loss": 0.7361,
+      "step": 490
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0002787037037037037,
+      "loss": 0.7663,
+      "step": 495
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00027839506172839504,
+      "loss": 0.7532,
+      "step": 500
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0002780864197530864,
+      "loss": 0.7573,
+      "step": 505
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0002777777777777778,
+      "loss": 0.7802,
+      "step": 510
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0002774691358024691,
+      "loss": 0.7871,
+      "step": 515
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00027716049382716046,
+      "loss": 0.7734,
+      "step": 520
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0002768518518518518,
+      "loss": 0.7752,
+      "step": 525
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0002765432098765432,
+      "loss": 0.7697,
+      "step": 530
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00027623456790123454,
+      "loss": 0.7846,
+      "step": 535
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0002759259259259259,
+      "loss": 0.7592,
+      "step": 540
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0002756172839506173,
+      "loss": 0.7734,
+      "step": 545
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0002753086419753086,
+      "loss": 0.7984,
+      "step": 550
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00027499999999999996,
+      "loss": 0.7342,
+      "step": 555
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00027469135802469136,
+      "loss": 0.7479,
+      "step": 560
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0002743827160493827,
+      "loss": 0.819,
+      "step": 565
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00027407407407407404,
+      "loss": 0.7561,
+      "step": 570
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0002737654320987654,
+      "loss": 0.7303,
+      "step": 575
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0002734567901234568,
+      "loss": 0.8001,
+      "step": 580
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0002731481481481481,
+      "loss": 0.7738,
+      "step": 585
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00027283950617283946,
+      "loss": 0.7866,
+      "step": 590
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00027253086419753086,
+      "loss": 0.7582,
+      "step": 595
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0002722222222222222,
+      "loss": 0.7645,
+      "step": 600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00027191358024691354,
+      "loss": 0.7839,
+      "step": 605
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0002716049382716049,
+      "loss": 0.7931,
+      "step": 610
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0002712962962962963,
+      "loss": 0.7692,
+      "step": 615
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0002709876543209876,
+      "loss": 0.7835,
+      "step": 620
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00027067901234567896,
+      "loss": 0.7795,
+      "step": 625
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00027037037037037036,
+      "loss": 0.7639,
+      "step": 630
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0002700617283950617,
+      "loss": 0.7705,
+      "step": 635
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00026975308641975304,
+      "loss": 0.7674,
+      "step": 640
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00026944444444444444,
+      "loss": 0.7349,
+      "step": 645
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0002691358024691358,
+      "loss": 0.8008,
+      "step": 650
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0002688271604938271,
+      "loss": 0.755,
+      "step": 655
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00026851851851851847,
+      "loss": 0.7526,
+      "step": 660
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00026820987654320986,
+      "loss": 0.7611,
+      "step": 665
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0002679012345679012,
+      "loss": 0.7621,
+      "step": 670
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00026759259259259255,
+      "loss": 0.7768,
+      "step": 675
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00026728395061728394,
+      "loss": 0.7703,
+      "step": 680
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0002669753086419753,
+      "loss": 0.7667,
+      "step": 685
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0002666666666666666,
+      "loss": 0.7712,
+      "step": 690
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.000266358024691358,
+      "loss": 0.7269,
+      "step": 695
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00026604938271604936,
+      "loss": 0.7275,
+      "step": 700
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0002657407407407407,
+      "loss": 0.7786,
+      "step": 705
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00026543209876543205,
+      "loss": 0.7808,
+      "step": 710
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00026512345679012344,
+      "loss": 0.7455,
+      "step": 715
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0002648148148148148,
+      "loss": 0.7717,
+      "step": 720
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0002645061728395061,
+      "loss": 0.737,
+      "step": 725
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0002641975308641975,
+      "loss": 0.7664,
+      "step": 730
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00026388888888888886,
+      "loss": 0.7716,
+      "step": 735
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00026358024691358026,
+      "loss": 0.7638,
+      "step": 740
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00026327160493827155,
+      "loss": 0.7777,
+      "step": 745
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00026296296296296294,
+      "loss": 0.7431,
+      "step": 750
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0002626543209876543,
+      "loss": 0.7516,
+      "step": 755
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00026234567901234563,
+      "loss": 0.831,
+      "step": 760
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.000262037037037037,
+      "loss": 0.7423,
+      "step": 765
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00026172839506172837,
+      "loss": 0.7471,
+      "step": 770
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0002614197530864197,
+      "loss": 0.7551,
+      "step": 775
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0002611111111111111,
+      "loss": 0.7208,
+      "step": 780
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00026080246913580245,
+      "loss": 0.7359,
+      "step": 785
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00026049382716049384,
+      "loss": 0.7759,
+      "step": 790
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00026018518518518513,
+      "loss": 0.7406,
+      "step": 795
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0002598765432098765,
+      "loss": 0.756,
+      "step": 800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00025956790123456787,
+      "loss": 0.7631,
+      "step": 805
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0002592592592592592,
+      "loss": 0.7885,
+      "step": 810
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0002589506172839506,
+      "loss": 0.7947,
+      "step": 815
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00025864197530864195,
+      "loss": 0.7557,
+      "step": 820
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00025833333333333334,
+      "loss": 0.7556,
+      "step": 825
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0002580246913580247,
+      "loss": 0.7472,
+      "step": 830
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.000257716049382716,
+      "loss": 0.729,
+      "step": 835
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0002574074074074074,
+      "loss": 0.7523,
+      "step": 840
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0002570987654320987,
+      "loss": 0.7505,
+      "step": 845
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0002567901234567901,
+      "loss": 0.7617,
+      "step": 850
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00025648148148148145,
+      "loss": 0.7898,
+      "step": 855
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00025617283950617284,
+      "loss": 0.7575,
+      "step": 860
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0002558641975308642,
+      "loss": 0.8316,
+      "step": 865
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00025555555555555553,
+      "loss": 0.7261,
+      "step": 870
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0002552469135802469,
+      "loss": 0.7594,
+      "step": 875
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00025493827160493826,
+      "loss": 0.7372,
+      "step": 880
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0002546296296296296,
+      "loss": 0.7498,
+      "step": 885
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00025432098765432095,
+      "loss": 0.7398,
+      "step": 890
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0002540123456790123,
+      "loss": 0.804,
+      "step": 895
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0002537037037037037,
+      "loss": 0.7241,
+      "step": 900
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00025339506172839503,
+      "loss": 0.7519,
+      "step": 905
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0002530864197530864,
+      "loss": 0.7694,
+      "step": 910
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00025277777777777777,
+      "loss": 0.7474,
+      "step": 915
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0002524691358024691,
+      "loss": 0.7278,
+      "step": 920
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0002521604938271605,
+      "loss": 0.7747,
+      "step": 925
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0002518518518518518,
+      "loss": 0.7858,
+      "step": 930
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0002515432098765432,
+      "loss": 0.7442,
+      "step": 935
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00025123456790123453,
+      "loss": 0.7446,
+      "step": 940
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0002509259259259259,
+      "loss": 0.7843,
+      "step": 945
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00025061728395061727,
+      "loss": 0.7512,
+      "step": 950
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0002503086419753086,
+      "loss": 0.7616,
+      "step": 955
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00025,
+      "loss": 0.826,
+      "step": 960
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00024969135802469135,
+      "loss": 0.7567,
+      "step": 965
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0002493827160493827,
+      "loss": 0.7754,
+      "step": 970
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0002490740740740741,
+      "loss": 0.7605,
+      "step": 975
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00024876543209876537,
+      "loss": 0.7728,
+      "step": 980
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00024845679012345677,
+      "loss": 0.7902,
+      "step": 985
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0002481481481481481,
+      "loss": 0.7622,
+      "step": 990
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0002478395061728395,
+      "loss": 0.745,
+      "step": 995
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00024753086419753085,
+      "loss": 0.7529,
+      "step": 1000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0002472222222222222,
+      "loss": 0.759,
+      "step": 1005
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0002469135802469136,
+      "loss": 0.7471,
+      "step": 1010
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00024660493827160493,
+      "loss": 0.7517,
+      "step": 1015
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00024629629629629627,
+      "loss": 0.7634,
+      "step": 1020
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002459876543209876,
+      "loss": 0.7833,
+      "step": 1025
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.000245679012345679,
+      "loss": 0.7227,
+      "step": 1030
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00024537037037037035,
+      "loss": 0.7792,
+      "step": 1035
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002450617283950617,
+      "loss": 0.7292,
+      "step": 1040
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002447530864197531,
+      "loss": 0.7519,
+      "step": 1045
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00024444444444444443,
+      "loss": 0.7385,
+      "step": 1050
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00024413580246913577,
+      "loss": 0.7842,
+      "step": 1055
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00024382716049382714,
+      "loss": 0.796,
+      "step": 1060
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002435185185185185,
+      "loss": 0.7479,
+      "step": 1065
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00024320987654320985,
+      "loss": 0.7531,
+      "step": 1070
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00024290123456790122,
+      "loss": 0.731,
+      "step": 1075
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002425925925925926,
+      "loss": 0.7319,
+      "step": 1080
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00024228395061728393,
+      "loss": 0.7381,
+      "step": 1085
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024197530864197527,
+      "loss": 0.7621,
+      "step": 1090
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024166666666666664,
+      "loss": 0.771,
+      "step": 1095
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024135802469135798,
+      "loss": 0.7493,
+      "step": 1100
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024104938271604935,
+      "loss": 0.7553,
+      "step": 1105
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024074074074074072,
+      "loss": 0.7862,
+      "step": 1110
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002404320987654321,
+      "loss": 0.7531,
+      "step": 1115
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00024012345679012343,
+      "loss": 0.7688,
+      "step": 1120
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002398148148148148,
+      "loss": 0.7666,
+      "step": 1125
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00023950617283950617,
+      "loss": 0.7484,
+      "step": 1130
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00023919753086419748,
+      "loss": 0.8048,
+      "step": 1135
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00023888888888888885,
+      "loss": 0.757,
+      "step": 1140
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00023858024691358022,
+      "loss": 0.7838,
+      "step": 1145
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002382716049382716,
+      "loss": 0.7834,
+      "step": 1150
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00023796296296296293,
+      "loss": 0.733,
+      "step": 1155
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002376543209876543,
+      "loss": 0.7195,
+      "step": 1160
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00023734567901234567,
+      "loss": 0.7478,
+      "step": 1165
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.000237037037037037,
+      "loss": 0.7456,
+      "step": 1170
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00023672839506172838,
+      "loss": 0.722,
+      "step": 1175
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00023641975308641975,
+      "loss": 0.7395,
+      "step": 1180
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00023611111111111112,
+      "loss": 0.7553,
+      "step": 1185
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00023580246913580243,
+      "loss": 0.7692,
+      "step": 1190
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002354938271604938,
+      "loss": 0.7396,
+      "step": 1195
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00023518518518518517,
+      "loss": 0.7214,
+      "step": 1200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002348765432098765,
+      "loss": 0.7359,
+      "step": 1205
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00023456790123456788,
+      "loss": 0.7666,
+      "step": 1210
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00023425925925925925,
+      "loss": 0.7583,
+      "step": 1215
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002339506172839506,
+      "loss": 0.7715,
+      "step": 1220
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00023364197530864196,
+      "loss": 0.7423,
+      "step": 1225
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002333333333333333,
+      "loss": 0.7598,
+      "step": 1230
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00023302469135802467,
+      "loss": 0.7251,
+      "step": 1235
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00023271604938271601,
+      "loss": 0.7447,
+      "step": 1240
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00023240740740740738,
+      "loss": 0.7923,
+      "step": 1245
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00023209876543209875,
+      "loss": 0.7161,
+      "step": 1250
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002317901234567901,
+      "loss": 0.7291,
+      "step": 1255
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00023148148148148146,
+      "loss": 0.7366,
+      "step": 1260
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00023117283950617283,
+      "loss": 0.7173,
+      "step": 1265
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002308641975308642,
+      "loss": 0.7575,
+      "step": 1270
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00023055555555555552,
+      "loss": 0.7632,
+      "step": 1275
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00023024691358024688,
+      "loss": 0.7601,
+      "step": 1280
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00022993827160493825,
+      "loss": 0.7395,
+      "step": 1285
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002296296296296296,
+      "loss": 0.7686,
+      "step": 1290
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00022932098765432096,
+      "loss": 0.7531,
+      "step": 1295
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00022901234567901233,
+      "loss": 0.7675,
+      "step": 1300
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002287037037037037,
+      "loss": 0.8103,
+      "step": 1305
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00022839506172839504,
+      "loss": 0.773,
+      "step": 1310
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002280864197530864,
+      "loss": 0.7549,
+      "step": 1315
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00022777777777777778,
+      "loss": 0.7547,
+      "step": 1320
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002274691358024691,
+      "loss": 0.758,
+      "step": 1325
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00022716049382716047,
+      "loss": 0.77,
+      "step": 1330
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00022685185185185183,
+      "loss": 0.7662,
+      "step": 1335
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00022654320987654318,
+      "loss": 0.744,
+      "step": 1340
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022623456790123455,
+      "loss": 0.7486,
+      "step": 1345
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022592592592592591,
+      "loss": 0.7473,
+      "step": 1350
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022561728395061728,
+      "loss": 0.7486,
+      "step": 1355
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022530864197530862,
+      "loss": 0.7205,
+      "step": 1360
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.000225,
+      "loss": 0.7289,
+      "step": 1365
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022469135802469134,
+      "loss": 0.7586,
+      "step": 1370
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022438271604938268,
+      "loss": 0.7175,
+      "step": 1375
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022407407407407405,
+      "loss": 0.6961,
+      "step": 1380
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022376543209876542,
+      "loss": 0.7242,
+      "step": 1385
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022345679012345678,
+      "loss": 0.7515,
+      "step": 1390
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022314814814814813,
+      "loss": 0.7522,
+      "step": 1395
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002228395061728395,
+      "loss": 0.7651,
+      "step": 1400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022253086419753086,
+      "loss": 0.6958,
+      "step": 1405
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022222222222222218,
+      "loss": 0.7336,
+      "step": 1410
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022191358024691355,
+      "loss": 0.7636,
+      "step": 1415
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00022160493827160492,
+      "loss": 0.7583,
+      "step": 1420
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00022129629629629626,
+      "loss": 0.7605,
+      "step": 1425
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00022098765432098763,
+      "loss": 0.7109,
+      "step": 1430
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.000220679012345679,
+      "loss": 0.7224,
+      "step": 1435
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00022037037037037037,
+      "loss": 0.7724,
+      "step": 1440
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002200617283950617,
+      "loss": 0.7637,
+      "step": 1445
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021975308641975308,
+      "loss": 0.7302,
+      "step": 1450
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021944444444444444,
+      "loss": 0.7685,
+      "step": 1455
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021913580246913576,
+      "loss": 0.751,
+      "step": 1460
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021882716049382713,
+      "loss": 0.7768,
+      "step": 1465
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002185185185185185,
+      "loss": 0.7461,
+      "step": 1470
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00021820987654320987,
+      "loss": 0.7782,
+      "step": 1475
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002179012345679012,
+      "loss": 0.7474,
+      "step": 1480
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00021759259259259258,
+      "loss": 0.7475,
+      "step": 1485
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00021728395061728395,
+      "loss": 0.7539,
+      "step": 1490
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002169753086419753,
+      "loss": 0.7476,
+      "step": 1495
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00021666666666666666,
+      "loss": 0.7382,
+      "step": 1500
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.000216358024691358,
+      "loss": 0.7305,
+      "step": 1505
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00021604938271604937,
+      "loss": 0.7455,
+      "step": 1510
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002157407407407407,
+      "loss": 0.7286,
+      "step": 1515
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00021543209876543208,
+      "loss": 0.7653,
+      "step": 1520
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00021512345679012345,
+      "loss": 0.7443,
+      "step": 1525
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0002148148148148148,
+      "loss": 0.7026,
+      "step": 1530
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00021450617283950616,
+      "loss": 0.7436,
+      "step": 1535
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00021419753086419753,
+      "loss": 0.7101,
+      "step": 1540
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00021388888888888884,
+      "loss": 0.7484,
+      "step": 1545
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002135802469135802,
+      "loss": 0.7483,
+      "step": 1550
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00021327160493827158,
+      "loss": 0.7419,
+      "step": 1555
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00021296296296296295,
+      "loss": 0.7356,
+      "step": 1560
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002126543209876543,
+      "loss": 0.7716,
+      "step": 1565
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00021234567901234566,
+      "loss": 0.734,
+      "step": 1570
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00021203703703703703,
+      "loss": 0.7268,
+      "step": 1575
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00021172839506172837,
+      "loss": 0.7133,
+      "step": 1580
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00021141975308641974,
+      "loss": 0.7575,
+      "step": 1585
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002111111111111111,
+      "loss": 0.7219,
+      "step": 1590
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00021080246913580248,
+      "loss": 0.7394,
+      "step": 1595
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0002104938271604938,
+      "loss": 0.7729,
+      "step": 1600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00021018518518518516,
+      "loss": 0.757,
+      "step": 1605
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00020987654320987653,
+      "loss": 0.7139,
+      "step": 1610
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00020956790123456787,
+      "loss": 0.7492,
+      "step": 1615
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00020925925925925924,
+      "loss": 0.7461,
+      "step": 1620
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002089506172839506,
+      "loss": 0.7276,
+      "step": 1625
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00020864197530864198,
+      "loss": 0.7725,
+      "step": 1630
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00020833333333333332,
+      "loss": 0.7623,
+      "step": 1635
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002080246913580247,
+      "loss": 0.7394,
+      "step": 1640
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00020771604938271603,
+      "loss": 0.7387,
+      "step": 1645
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00020740740740740737,
+      "loss": 0.7519,
+      "step": 1650
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00020709876543209874,
+      "loss": 0.7625,
+      "step": 1655
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002067901234567901,
+      "loss": 0.7254,
+      "step": 1660
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00020648148148148145,
+      "loss": 0.7129,
+      "step": 1665
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00020617283950617282,
+      "loss": 0.7964,
+      "step": 1670
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002058641975308642,
+      "loss": 0.7332,
+      "step": 1675
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00020555555555555556,
+      "loss": 0.7955,
+      "step": 1680
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00020524691358024687,
+      "loss": 0.7544,
+      "step": 1685
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00020493827160493824,
+      "loss": 0.7115,
+      "step": 1690
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002046296296296296,
+      "loss": 0.7434,
+      "step": 1695
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00020432098765432095,
+      "loss": 0.7382,
+      "step": 1700
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00020401234567901232,
+      "loss": 0.7452,
+      "step": 1705
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002037037037037037,
+      "loss": 0.7551,
+      "step": 1710
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00020339506172839506,
+      "loss": 0.7202,
+      "step": 1715
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002030864197530864,
+      "loss": 0.735,
+      "step": 1720
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00020277777777777777,
+      "loss": 0.7581,
+      "step": 1725
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00020246913580246914,
+      "loss": 0.7527,
+      "step": 1730
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00020216049382716045,
+      "loss": 0.7498,
+      "step": 1735
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00020185185185185182,
+      "loss": 0.735,
+      "step": 1740
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002015432098765432,
+      "loss": 0.7286,
+      "step": 1745
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00020123456790123456,
+      "loss": 0.6891,
+      "step": 1750
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002009259259259259,
+      "loss": 0.7353,
+      "step": 1755
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00020061728395061727,
+      "loss": 0.7386,
+      "step": 1760
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00020030864197530864,
+      "loss": 0.7106,
+      "step": 1765
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00019999999999999998,
+      "loss": 0.7207,
+      "step": 1770
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00019969135802469135,
+      "loss": 0.7576,
+      "step": 1775
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00019938271604938272,
+      "loss": 0.7462,
+      "step": 1780
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00019907407407407404,
+      "loss": 0.7068,
+      "step": 1785
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001987654320987654,
+      "loss": 0.7525,
+      "step": 1790
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019845679012345677,
+      "loss": 0.7291,
+      "step": 1795
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019814814814814814,
+      "loss": 0.7226,
+      "step": 1800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019783950617283948,
+      "loss": 0.7445,
+      "step": 1805
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019753086419753085,
+      "loss": 0.7571,
+      "step": 1810
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019722222222222222,
+      "loss": 0.7306,
+      "step": 1815
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00019691358024691354,
+      "loss": 0.7785,
+      "step": 1820
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001966049382716049,
+      "loss": 0.7537,
+      "step": 1825
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00019629629629629627,
+      "loss": 0.7549,
+      "step": 1830
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00019598765432098764,
+      "loss": 0.732,
+      "step": 1835
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00019567901234567898,
+      "loss": 0.7159,
+      "step": 1840
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00019537037037037035,
+      "loss": 0.7549,
+      "step": 1845
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00019506172839506172,
+      "loss": 0.7581,
+      "step": 1850
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00019475308641975306,
+      "loss": 0.7559,
+      "step": 1855
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00019444444444444443,
+      "loss": 0.7626,
+      "step": 1860
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001941358024691358,
+      "loss": 0.7423,
+      "step": 1865
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019382716049382712,
+      "loss": 0.7672,
+      "step": 1870
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019351851851851849,
+      "loss": 0.7794,
+      "step": 1875
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019320987654320985,
+      "loss": 0.7428,
+      "step": 1880
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019290123456790122,
+      "loss": 0.7462,
+      "step": 1885
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019259259259259257,
+      "loss": 0.702,
+      "step": 1890
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00019228395061728393,
+      "loss": 0.7626,
+      "step": 1895
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001919753086419753,
+      "loss": 0.7357,
+      "step": 1900
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00019166666666666665,
+      "loss": 0.7408,
+      "step": 1905
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00019135802469135801,
+      "loss": 0.707,
+      "step": 1910
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00019104938271604938,
+      "loss": 0.726,
+      "step": 1915
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00019074074074074073,
+      "loss": 0.7396,
+      "step": 1920
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00019043209876543207,
+      "loss": 0.7676,
+      "step": 1925
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00019012345679012344,
+      "loss": 0.7335,
+      "step": 1930
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0001898148148148148,
+      "loss": 0.7235,
+      "step": 1935
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00018950617283950615,
+      "loss": 0.7297,
+      "step": 1940
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018919753086419752,
+      "loss": 0.7079,
+      "step": 1945
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018888888888888888,
+      "loss": 0.7629,
+      "step": 1950
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018858024691358025,
+      "loss": 0.7131,
+      "step": 1955
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018827160493827157,
+      "loss": 0.7517,
+      "step": 1960
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018796296296296294,
+      "loss": 0.7292,
+      "step": 1965
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001876543209876543,
+      "loss": 0.7137,
+      "step": 1970
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00018734567901234565,
+      "loss": 0.725,
+      "step": 1975
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00018703703703703702,
+      "loss": 0.7444,
+      "step": 1980
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00018672839506172839,
+      "loss": 0.7366,
+      "step": 1985
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00018641975308641973,
+      "loss": 0.7329,
+      "step": 1990
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0001861111111111111,
+      "loss": 0.7037,
+      "step": 1995
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00018580246913580247,
+      "loss": 0.7186,
+      "step": 2000
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 5010,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 500,
+  "total_flos": 8.056468541093708e+18,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

save_lora_instruct_r8_3e4_s11_1/checkpoint-2000/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6bc4247ac8c6f92ef49080d32b2520dd809daff76a4805fdafb6e454ee1c5cc3
+size 5944

save_lora_instruct_r8_3e4_s11_1/checkpoint-2000/zero_to_fp32.py

@@ -0,0 +1,587 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel()
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)

save_lora_instruct_r8_3e4_s11_1/checkpoint-2500/README.md

@@ -0,0 +1,202 @@
+---
+base_model: meta-llama/Llama-2-7b-hf
+library_name: peft
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.9.0

save_lora_instruct_r8_3e4_s11_1/checkpoint-2500/adapter_config.json

@@ -0,0 +1,35 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "meta-llama/Llama-2-7b-hf",
+  "bias": "none",
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 16,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "mora_type": 1,
+  "peft_type": "LORA",
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "k_proj",
+    "v_proj",
+    "o_proj",
+    "gate_proj",
+    "down_proj",
+    "up_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_mora": false,
+  "use_rslora": false
+}

save_lora_instruct_r8_3e4_s11_1/checkpoint-2500/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:643f61bad9db98135ceda0a6687ae04402e2a9ee74aef71abae3ef8d4469ebe9
+size 564342240

save_lora_instruct_r8_3e4_s11_1/checkpoint-2500/global_step2500/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ab7725cf6a626b65267ae4e798d53e18d6e34318adc63460b4eb762c59f1ebb7
+size 239914224

save_lora_instruct_r8_3e4_s11_1/checkpoint-2500/global_step2500/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f5b28e910a452756e2ab041879d53daf14e0f79f61b8d9a2db7dda25a69d8164
+size 40303404

save_lora_instruct_r8_3e4_s11_1/checkpoint-2500/latest

@@ -0,0 +1 @@
+global_step2500

save_lora_instruct_r8_3e4_s11_1/checkpoint-2500/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f825f29d46d68de12985e875822b27efaac55bdba6c419fb26c2f1eb068aef1
+size 14244

save_lora_instruct_r8_3e4_s11_1/checkpoint-2500/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c65c8c33b44f669cd4e4fc53ad8a403f29e98144aa664322f23d1df8c3e22cd6
+size 1064

save_lora_instruct_r8_3e4_s11_1/checkpoint-2500/trainer_state.json

@@ -0,0 +1,3021 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.9979106246296816,
+  "eval_steps": 500,
+  "global_step": 2500,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.999999999999999e-06,
+      "loss": 1.38,
+      "step": 5
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 1.9999999999999998e-05,
+      "loss": 1.3936,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 2.9999999999999997e-05,
+      "loss": 1.3006,
+      "step": 15
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.9999999999999996e-05,
+      "loss": 1.2594,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 4.9999999999999996e-05,
+      "loss": 1.1177,
+      "step": 25
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 5.9999999999999995e-05,
+      "loss": 1.0481,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 7e-05,
+      "loss": 1.0601,
+      "step": 35
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 7.999999999999999e-05,
+      "loss": 1.0473,
+      "step": 40
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 8.999999999999999e-05,
+      "loss": 0.9523,
+      "step": 45
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.999999999999999e-05,
+      "loss": 0.9451,
+      "step": 50
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00010999999999999998,
+      "loss": 0.9687,
+      "step": 55
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00011999999999999999,
+      "loss": 0.9054,
+      "step": 60
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00013,
+      "loss": 0.9518,
+      "step": 65
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00014,
+      "loss": 0.8763,
+      "step": 70
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00015,
+      "loss": 0.8657,
+      "step": 75
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00015999999999999999,
+      "loss": 0.9145,
+      "step": 80
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00016999999999999999,
+      "loss": 0.8296,
+      "step": 85
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00017999999999999998,
+      "loss": 0.8709,
+      "step": 90
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00018999999999999998,
+      "loss": 0.8405,
+      "step": 95
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00019999999999999998,
+      "loss": 0.8604,
+      "step": 100
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00020999999999999998,
+      "loss": 0.8607,
+      "step": 105
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00021999999999999995,
+      "loss": 0.8376,
+      "step": 110
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00023,
+      "loss": 0.8554,
+      "step": 115
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00023999999999999998,
+      "loss": 0.8456,
+      "step": 120
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00025,
+      "loss": 0.8443,
+      "step": 125
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00026,
+      "loss": 0.8327,
+      "step": 130
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00027,
+      "loss": 0.8096,
+      "step": 135
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00028,
+      "loss": 0.7972,
+      "step": 140
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00029,
+      "loss": 0.8306,
+      "step": 145
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0003,
+      "loss": 0.8409,
+      "step": 150
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0002996913580246913,
+      "loss": 0.8094,
+      "step": 155
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00029938271604938266,
+      "loss": 0.8146,
+      "step": 160
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00029907407407407405,
+      "loss": 0.8042,
+      "step": 165
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0002987654320987654,
+      "loss": 0.7784,
+      "step": 170
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0002984567901234568,
+      "loss": 0.828,
+      "step": 175
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00029814814814814813,
+      "loss": 0.8473,
+      "step": 180
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0002978395061728395,
+      "loss": 0.8354,
+      "step": 185
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00029753086419753087,
+      "loss": 0.8246,
+      "step": 190
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00029722222222222216,
+      "loss": 0.8249,
+      "step": 195
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00029691358024691355,
+      "loss": 0.8222,
+      "step": 200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0002966049382716049,
+      "loss": 0.8369,
+      "step": 205
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00029629629629629624,
+      "loss": 0.7954,
+      "step": 210
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00029598765432098763,
+      "loss": 0.8051,
+      "step": 215
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000295679012345679,
+      "loss": 0.8081,
+      "step": 220
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00029537037037037037,
+      "loss": 0.7864,
+      "step": 225
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0002950617283950617,
+      "loss": 0.7974,
+      "step": 230
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00029475308641975306,
+      "loss": 0.8306,
+      "step": 235
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00029444444444444445,
+      "loss": 0.7792,
+      "step": 240
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00029413580246913574,
+      "loss": 0.8108,
+      "step": 245
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00029382716049382714,
+      "loss": 0.8167,
+      "step": 250
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0002935185185185185,
+      "loss": 0.8358,
+      "step": 255
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0002932098765432099,
+      "loss": 0.7948,
+      "step": 260
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0002929012345679012,
+      "loss": 0.8078,
+      "step": 265
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00029259259259259256,
+      "loss": 0.7886,
+      "step": 270
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00029228395061728395,
+      "loss": 0.7903,
+      "step": 275
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0002919753086419753,
+      "loss": 0.7892,
+      "step": 280
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00029166666666666664,
+      "loss": 0.7717,
+      "step": 285
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.000291358024691358,
+      "loss": 0.7774,
+      "step": 290
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0002910493827160494,
+      "loss": 0.8053,
+      "step": 295
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0002907407407407407,
+      "loss": 0.794,
+      "step": 300
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00029043209876543206,
+      "loss": 0.7856,
+      "step": 305
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00029012345679012345,
+      "loss": 0.7713,
+      "step": 310
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0002898148148148148,
+      "loss": 0.7805,
+      "step": 315
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00028950617283950614,
+      "loss": 0.7897,
+      "step": 320
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00028919753086419753,
+      "loss": 0.7931,
+      "step": 325
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0002888888888888888,
+      "loss": 0.7739,
+      "step": 330
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0002885802469135802,
+      "loss": 0.7749,
+      "step": 335
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00028827160493827156,
+      "loss": 0.7932,
+      "step": 340
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00028796296296296296,
+      "loss": 0.7906,
+      "step": 345
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0002876543209876543,
+      "loss": 0.8094,
+      "step": 350
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00028734567901234564,
+      "loss": 0.7932,
+      "step": 355
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00028703703703703703,
+      "loss": 0.8027,
+      "step": 360
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0002867283950617284,
+      "loss": 0.8297,
+      "step": 365
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0002864197530864197,
+      "loss": 0.8228,
+      "step": 370
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0002861111111111111,
+      "loss": 0.7997,
+      "step": 375
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00028580246913580246,
+      "loss": 0.8124,
+      "step": 380
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0002854938271604938,
+      "loss": 0.8152,
+      "step": 385
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00028518518518518514,
+      "loss": 0.7833,
+      "step": 390
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00028487654320987654,
+      "loss": 0.7884,
+      "step": 395
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0002845679012345679,
+      "loss": 0.753,
+      "step": 400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0002842592592592592,
+      "loss": 0.8055,
+      "step": 405
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0002839506172839506,
+      "loss": 0.8014,
+      "step": 410
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00028364197530864196,
+      "loss": 0.782,
+      "step": 415
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0002833333333333333,
+      "loss": 0.7733,
+      "step": 420
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0002830246913580247,
+      "loss": 0.7777,
+      "step": 425
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00028271604938271604,
+      "loss": 0.752,
+      "step": 430
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0002824074074074074,
+      "loss": 0.7963,
+      "step": 435
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0002820987654320987,
+      "loss": 0.7655,
+      "step": 440
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0002817901234567901,
+      "loss": 0.8,
+      "step": 445
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00028148148148148146,
+      "loss": 0.7465,
+      "step": 450
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0002811728395061728,
+      "loss": 0.7432,
+      "step": 455
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0002808641975308642,
+      "loss": 0.758,
+      "step": 460
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00028055555555555554,
+      "loss": 0.7757,
+      "step": 465
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0002802469135802469,
+      "loss": 0.7816,
+      "step": 470
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0002799382716049382,
+      "loss": 0.7744,
+      "step": 475
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0002796296296296296,
+      "loss": 0.7807,
+      "step": 480
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00027932098765432096,
+      "loss": 0.774,
+      "step": 485
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0002790123456790123,
+      "loss": 0.7361,
+      "step": 490
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0002787037037037037,
+      "loss": 0.7663,
+      "step": 495
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00027839506172839504,
+      "loss": 0.7532,
+      "step": 500
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0002780864197530864,
+      "loss": 0.7573,
+      "step": 505
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0002777777777777778,
+      "loss": 0.7802,
+      "step": 510
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0002774691358024691,
+      "loss": 0.7871,
+      "step": 515
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00027716049382716046,
+      "loss": 0.7734,
+      "step": 520
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0002768518518518518,
+      "loss": 0.7752,
+      "step": 525
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0002765432098765432,
+      "loss": 0.7697,
+      "step": 530
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00027623456790123454,
+      "loss": 0.7846,
+      "step": 535
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0002759259259259259,
+      "loss": 0.7592,
+      "step": 540
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0002756172839506173,
+      "loss": 0.7734,
+      "step": 545
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0002753086419753086,
+      "loss": 0.7984,
+      "step": 550
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00027499999999999996,
+      "loss": 0.7342,
+      "step": 555
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00027469135802469136,
+      "loss": 0.7479,
+      "step": 560
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0002743827160493827,
+      "loss": 0.819,
+      "step": 565
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00027407407407407404,
+      "loss": 0.7561,
+      "step": 570
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0002737654320987654,
+      "loss": 0.7303,
+      "step": 575
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0002734567901234568,
+      "loss": 0.8001,
+      "step": 580
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0002731481481481481,
+      "loss": 0.7738,
+      "step": 585
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00027283950617283946,
+      "loss": 0.7866,
+      "step": 590
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00027253086419753086,
+      "loss": 0.7582,
+      "step": 595
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0002722222222222222,
+      "loss": 0.7645,
+      "step": 600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00027191358024691354,
+      "loss": 0.7839,
+      "step": 605
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0002716049382716049,
+      "loss": 0.7931,
+      "step": 610
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0002712962962962963,
+      "loss": 0.7692,
+      "step": 615
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0002709876543209876,
+      "loss": 0.7835,
+      "step": 620
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00027067901234567896,
+      "loss": 0.7795,
+      "step": 625
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00027037037037037036,
+      "loss": 0.7639,
+      "step": 630
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0002700617283950617,
+      "loss": 0.7705,
+      "step": 635
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00026975308641975304,
+      "loss": 0.7674,
+      "step": 640
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00026944444444444444,
+      "loss": 0.7349,
+      "step": 645
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0002691358024691358,
+      "loss": 0.8008,
+      "step": 650
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0002688271604938271,
+      "loss": 0.755,
+      "step": 655
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00026851851851851847,
+      "loss": 0.7526,
+      "step": 660
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00026820987654320986,
+      "loss": 0.7611,
+      "step": 665
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0002679012345679012,
+      "loss": 0.7621,
+      "step": 670
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00026759259259259255,
+      "loss": 0.7768,
+      "step": 675
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00026728395061728394,
+      "loss": 0.7703,
+      "step": 680
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0002669753086419753,
+      "loss": 0.7667,
+      "step": 685
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0002666666666666666,
+      "loss": 0.7712,
+      "step": 690
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.000266358024691358,
+      "loss": 0.7269,
+      "step": 695
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00026604938271604936,
+      "loss": 0.7275,
+      "step": 700
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0002657407407407407,
+      "loss": 0.7786,
+      "step": 705
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00026543209876543205,
+      "loss": 0.7808,
+      "step": 710
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00026512345679012344,
+      "loss": 0.7455,
+      "step": 715
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0002648148148148148,
+      "loss": 0.7717,
+      "step": 720
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0002645061728395061,
+      "loss": 0.737,
+      "step": 725
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0002641975308641975,
+      "loss": 0.7664,
+      "step": 730
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00026388888888888886,
+      "loss": 0.7716,
+      "step": 735
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00026358024691358026,
+      "loss": 0.7638,
+      "step": 740
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00026327160493827155,
+      "loss": 0.7777,
+      "step": 745
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00026296296296296294,
+      "loss": 0.7431,
+      "step": 750
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0002626543209876543,
+      "loss": 0.7516,
+      "step": 755
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00026234567901234563,
+      "loss": 0.831,
+      "step": 760
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.000262037037037037,
+      "loss": 0.7423,
+      "step": 765
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00026172839506172837,
+      "loss": 0.7471,
+      "step": 770
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0002614197530864197,
+      "loss": 0.7551,
+      "step": 775
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0002611111111111111,
+      "loss": 0.7208,
+      "step": 780
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00026080246913580245,
+      "loss": 0.7359,
+      "step": 785
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00026049382716049384,
+      "loss": 0.7759,
+      "step": 790
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00026018518518518513,
+      "loss": 0.7406,
+      "step": 795
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0002598765432098765,
+      "loss": 0.756,
+      "step": 800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00025956790123456787,
+      "loss": 0.7631,
+      "step": 805
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0002592592592592592,
+      "loss": 0.7885,
+      "step": 810
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0002589506172839506,
+      "loss": 0.7947,
+      "step": 815
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00025864197530864195,
+      "loss": 0.7557,
+      "step": 820
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00025833333333333334,
+      "loss": 0.7556,
+      "step": 825
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0002580246913580247,
+      "loss": 0.7472,
+      "step": 830
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.000257716049382716,
+      "loss": 0.729,
+      "step": 835
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0002574074074074074,
+      "loss": 0.7523,
+      "step": 840
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0002570987654320987,
+      "loss": 0.7505,
+      "step": 845
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0002567901234567901,
+      "loss": 0.7617,
+      "step": 850
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00025648148148148145,
+      "loss": 0.7898,
+      "step": 855
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00025617283950617284,
+      "loss": 0.7575,
+      "step": 860
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0002558641975308642,
+      "loss": 0.8316,
+      "step": 865
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00025555555555555553,
+      "loss": 0.7261,
+      "step": 870
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0002552469135802469,
+      "loss": 0.7594,
+      "step": 875
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00025493827160493826,
+      "loss": 0.7372,
+      "step": 880
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0002546296296296296,
+      "loss": 0.7498,
+      "step": 885
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00025432098765432095,
+      "loss": 0.7398,
+      "step": 890
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0002540123456790123,
+      "loss": 0.804,
+      "step": 895
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0002537037037037037,
+      "loss": 0.7241,
+      "step": 900
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00025339506172839503,
+      "loss": 0.7519,
+      "step": 905
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0002530864197530864,
+      "loss": 0.7694,
+      "step": 910
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00025277777777777777,
+      "loss": 0.7474,
+      "step": 915
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0002524691358024691,
+      "loss": 0.7278,
+      "step": 920
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0002521604938271605,
+      "loss": 0.7747,
+      "step": 925
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0002518518518518518,
+      "loss": 0.7858,
+      "step": 930
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0002515432098765432,
+      "loss": 0.7442,
+      "step": 935
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00025123456790123453,
+      "loss": 0.7446,
+      "step": 940
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0002509259259259259,
+      "loss": 0.7843,
+      "step": 945
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00025061728395061727,
+      "loss": 0.7512,
+      "step": 950
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0002503086419753086,
+      "loss": 0.7616,
+      "step": 955
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00025,
+      "loss": 0.826,
+      "step": 960
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00024969135802469135,
+      "loss": 0.7567,
+      "step": 965
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0002493827160493827,
+      "loss": 0.7754,
+      "step": 970
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0002490740740740741,
+      "loss": 0.7605,
+      "step": 975
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00024876543209876537,
+      "loss": 0.7728,
+      "step": 980
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00024845679012345677,
+      "loss": 0.7902,
+      "step": 985
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0002481481481481481,
+      "loss": 0.7622,
+      "step": 990
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0002478395061728395,
+      "loss": 0.745,
+      "step": 995
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00024753086419753085,
+      "loss": 0.7529,
+      "step": 1000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0002472222222222222,
+      "loss": 0.759,
+      "step": 1005
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0002469135802469136,
+      "loss": 0.7471,
+      "step": 1010
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00024660493827160493,
+      "loss": 0.7517,
+      "step": 1015
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00024629629629629627,
+      "loss": 0.7634,
+      "step": 1020
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002459876543209876,
+      "loss": 0.7833,
+      "step": 1025
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.000245679012345679,
+      "loss": 0.7227,
+      "step": 1030
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00024537037037037035,
+      "loss": 0.7792,
+      "step": 1035
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002450617283950617,
+      "loss": 0.7292,
+      "step": 1040
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002447530864197531,
+      "loss": 0.7519,
+      "step": 1045
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00024444444444444443,
+      "loss": 0.7385,
+      "step": 1050
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00024413580246913577,
+      "loss": 0.7842,
+      "step": 1055
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00024382716049382714,
+      "loss": 0.796,
+      "step": 1060
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002435185185185185,
+      "loss": 0.7479,
+      "step": 1065
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00024320987654320985,
+      "loss": 0.7531,
+      "step": 1070
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00024290123456790122,
+      "loss": 0.731,
+      "step": 1075
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002425925925925926,
+      "loss": 0.7319,
+      "step": 1080
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00024228395061728393,
+      "loss": 0.7381,
+      "step": 1085
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024197530864197527,
+      "loss": 0.7621,
+      "step": 1090
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024166666666666664,
+      "loss": 0.771,
+      "step": 1095
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024135802469135798,
+      "loss": 0.7493,
+      "step": 1100
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024104938271604935,
+      "loss": 0.7553,
+      "step": 1105
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024074074074074072,
+      "loss": 0.7862,
+      "step": 1110
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002404320987654321,
+      "loss": 0.7531,
+      "step": 1115
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00024012345679012343,
+      "loss": 0.7688,
+      "step": 1120
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002398148148148148,
+      "loss": 0.7666,
+      "step": 1125
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00023950617283950617,
+      "loss": 0.7484,
+      "step": 1130
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00023919753086419748,
+      "loss": 0.8048,
+      "step": 1135
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00023888888888888885,
+      "loss": 0.757,
+      "step": 1140
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00023858024691358022,
+      "loss": 0.7838,
+      "step": 1145
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002382716049382716,
+      "loss": 0.7834,
+      "step": 1150
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00023796296296296293,
+      "loss": 0.733,
+      "step": 1155
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002376543209876543,
+      "loss": 0.7195,
+      "step": 1160
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00023734567901234567,
+      "loss": 0.7478,
+      "step": 1165
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.000237037037037037,
+      "loss": 0.7456,
+      "step": 1170
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00023672839506172838,
+      "loss": 0.722,
+      "step": 1175
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00023641975308641975,
+      "loss": 0.7395,
+      "step": 1180
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00023611111111111112,
+      "loss": 0.7553,
+      "step": 1185
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00023580246913580243,
+      "loss": 0.7692,
+      "step": 1190
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002354938271604938,
+      "loss": 0.7396,
+      "step": 1195
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00023518518518518517,
+      "loss": 0.7214,
+      "step": 1200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002348765432098765,
+      "loss": 0.7359,
+      "step": 1205
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00023456790123456788,
+      "loss": 0.7666,
+      "step": 1210
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00023425925925925925,
+      "loss": 0.7583,
+      "step": 1215
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002339506172839506,
+      "loss": 0.7715,
+      "step": 1220
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00023364197530864196,
+      "loss": 0.7423,
+      "step": 1225
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002333333333333333,
+      "loss": 0.7598,
+      "step": 1230
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00023302469135802467,
+      "loss": 0.7251,
+      "step": 1235
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00023271604938271601,
+      "loss": 0.7447,
+      "step": 1240
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00023240740740740738,
+      "loss": 0.7923,
+      "step": 1245
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00023209876543209875,
+      "loss": 0.7161,
+      "step": 1250
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002317901234567901,
+      "loss": 0.7291,
+      "step": 1255
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00023148148148148146,
+      "loss": 0.7366,
+      "step": 1260
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00023117283950617283,
+      "loss": 0.7173,
+      "step": 1265
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002308641975308642,
+      "loss": 0.7575,
+      "step": 1270
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00023055555555555552,
+      "loss": 0.7632,
+      "step": 1275
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00023024691358024688,
+      "loss": 0.7601,
+      "step": 1280
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00022993827160493825,
+      "loss": 0.7395,
+      "step": 1285
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002296296296296296,
+      "loss": 0.7686,
+      "step": 1290
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00022932098765432096,
+      "loss": 0.7531,
+      "step": 1295
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00022901234567901233,
+      "loss": 0.7675,
+      "step": 1300
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002287037037037037,
+      "loss": 0.8103,
+      "step": 1305
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00022839506172839504,
+      "loss": 0.773,
+      "step": 1310
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002280864197530864,
+      "loss": 0.7549,
+      "step": 1315
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00022777777777777778,
+      "loss": 0.7547,
+      "step": 1320
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002274691358024691,
+      "loss": 0.758,
+      "step": 1325
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00022716049382716047,
+      "loss": 0.77,
+      "step": 1330
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00022685185185185183,
+      "loss": 0.7662,
+      "step": 1335
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00022654320987654318,
+      "loss": 0.744,
+      "step": 1340
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022623456790123455,
+      "loss": 0.7486,
+      "step": 1345
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022592592592592591,
+      "loss": 0.7473,
+      "step": 1350
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022561728395061728,
+      "loss": 0.7486,
+      "step": 1355
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022530864197530862,
+      "loss": 0.7205,
+      "step": 1360
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.000225,
+      "loss": 0.7289,
+      "step": 1365
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022469135802469134,
+      "loss": 0.7586,
+      "step": 1370
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022438271604938268,
+      "loss": 0.7175,
+      "step": 1375
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022407407407407405,
+      "loss": 0.6961,
+      "step": 1380
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022376543209876542,
+      "loss": 0.7242,
+      "step": 1385
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022345679012345678,
+      "loss": 0.7515,
+      "step": 1390
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022314814814814813,
+      "loss": 0.7522,
+      "step": 1395
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002228395061728395,
+      "loss": 0.7651,
+      "step": 1400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022253086419753086,
+      "loss": 0.6958,
+      "step": 1405
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022222222222222218,
+      "loss": 0.7336,
+      "step": 1410
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022191358024691355,
+      "loss": 0.7636,
+      "step": 1415
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00022160493827160492,
+      "loss": 0.7583,
+      "step": 1420
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00022129629629629626,
+      "loss": 0.7605,
+      "step": 1425
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00022098765432098763,
+      "loss": 0.7109,
+      "step": 1430
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.000220679012345679,
+      "loss": 0.7224,
+      "step": 1435
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00022037037037037037,
+      "loss": 0.7724,
+      "step": 1440
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002200617283950617,
+      "loss": 0.7637,
+      "step": 1445
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021975308641975308,
+      "loss": 0.7302,
+      "step": 1450
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021944444444444444,
+      "loss": 0.7685,
+      "step": 1455
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021913580246913576,
+      "loss": 0.751,
+      "step": 1460
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021882716049382713,
+      "loss": 0.7768,
+      "step": 1465
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002185185185185185,
+      "loss": 0.7461,
+      "step": 1470
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00021820987654320987,
+      "loss": 0.7782,
+      "step": 1475
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002179012345679012,
+      "loss": 0.7474,
+      "step": 1480
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00021759259259259258,
+      "loss": 0.7475,
+      "step": 1485
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00021728395061728395,
+      "loss": 0.7539,
+      "step": 1490
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002169753086419753,
+      "loss": 0.7476,
+      "step": 1495
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00021666666666666666,
+      "loss": 0.7382,
+      "step": 1500
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.000216358024691358,
+      "loss": 0.7305,
+      "step": 1505
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00021604938271604937,
+      "loss": 0.7455,
+      "step": 1510
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002157407407407407,
+      "loss": 0.7286,
+      "step": 1515
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00021543209876543208,
+      "loss": 0.7653,
+      "step": 1520
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00021512345679012345,
+      "loss": 0.7443,
+      "step": 1525
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0002148148148148148,
+      "loss": 0.7026,
+      "step": 1530
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00021450617283950616,
+      "loss": 0.7436,
+      "step": 1535
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00021419753086419753,
+      "loss": 0.7101,
+      "step": 1540
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00021388888888888884,
+      "loss": 0.7484,
+      "step": 1545
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002135802469135802,
+      "loss": 0.7483,
+      "step": 1550
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00021327160493827158,
+      "loss": 0.7419,
+      "step": 1555
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00021296296296296295,
+      "loss": 0.7356,
+      "step": 1560
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002126543209876543,
+      "loss": 0.7716,
+      "step": 1565
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00021234567901234566,
+      "loss": 0.734,
+      "step": 1570
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00021203703703703703,
+      "loss": 0.7268,
+      "step": 1575
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00021172839506172837,
+      "loss": 0.7133,
+      "step": 1580
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00021141975308641974,
+      "loss": 0.7575,
+      "step": 1585
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002111111111111111,
+      "loss": 0.7219,
+      "step": 1590
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00021080246913580248,
+      "loss": 0.7394,
+      "step": 1595
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0002104938271604938,
+      "loss": 0.7729,
+      "step": 1600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00021018518518518516,
+      "loss": 0.757,
+      "step": 1605
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00020987654320987653,
+      "loss": 0.7139,
+      "step": 1610
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00020956790123456787,
+      "loss": 0.7492,
+      "step": 1615
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00020925925925925924,
+      "loss": 0.7461,
+      "step": 1620
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002089506172839506,
+      "loss": 0.7276,
+      "step": 1625
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00020864197530864198,
+      "loss": 0.7725,
+      "step": 1630
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00020833333333333332,
+      "loss": 0.7623,
+      "step": 1635
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002080246913580247,
+      "loss": 0.7394,
+      "step": 1640
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00020771604938271603,
+      "loss": 0.7387,
+      "step": 1645
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00020740740740740737,
+      "loss": 0.7519,
+      "step": 1650
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00020709876543209874,
+      "loss": 0.7625,
+      "step": 1655
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002067901234567901,
+      "loss": 0.7254,
+      "step": 1660
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00020648148148148145,
+      "loss": 0.7129,
+      "step": 1665
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00020617283950617282,
+      "loss": 0.7964,
+      "step": 1670
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002058641975308642,
+      "loss": 0.7332,
+      "step": 1675
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00020555555555555556,
+      "loss": 0.7955,
+      "step": 1680
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00020524691358024687,
+      "loss": 0.7544,
+      "step": 1685
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00020493827160493824,
+      "loss": 0.7115,
+      "step": 1690
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002046296296296296,
+      "loss": 0.7434,
+      "step": 1695
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00020432098765432095,
+      "loss": 0.7382,
+      "step": 1700
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00020401234567901232,
+      "loss": 0.7452,
+      "step": 1705
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002037037037037037,
+      "loss": 0.7551,
+      "step": 1710
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00020339506172839506,
+      "loss": 0.7202,
+      "step": 1715
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002030864197530864,
+      "loss": 0.735,
+      "step": 1720
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00020277777777777777,
+      "loss": 0.7581,
+      "step": 1725
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00020246913580246914,
+      "loss": 0.7527,
+      "step": 1730
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00020216049382716045,
+      "loss": 0.7498,
+      "step": 1735
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00020185185185185182,
+      "loss": 0.735,
+      "step": 1740
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002015432098765432,
+      "loss": 0.7286,
+      "step": 1745
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00020123456790123456,
+      "loss": 0.6891,
+      "step": 1750
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002009259259259259,
+      "loss": 0.7353,
+      "step": 1755
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00020061728395061727,
+      "loss": 0.7386,
+      "step": 1760
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00020030864197530864,
+      "loss": 0.7106,
+      "step": 1765
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00019999999999999998,
+      "loss": 0.7207,
+      "step": 1770
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00019969135802469135,
+      "loss": 0.7576,
+      "step": 1775
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00019938271604938272,
+      "loss": 0.7462,
+      "step": 1780
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00019907407407407404,
+      "loss": 0.7068,
+      "step": 1785
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001987654320987654,
+      "loss": 0.7525,
+      "step": 1790
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019845679012345677,
+      "loss": 0.7291,
+      "step": 1795
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019814814814814814,
+      "loss": 0.7226,
+      "step": 1800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019783950617283948,
+      "loss": 0.7445,
+      "step": 1805
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019753086419753085,
+      "loss": 0.7571,
+      "step": 1810
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019722222222222222,
+      "loss": 0.7306,
+      "step": 1815
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00019691358024691354,
+      "loss": 0.7785,
+      "step": 1820
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001966049382716049,
+      "loss": 0.7537,
+      "step": 1825
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00019629629629629627,
+      "loss": 0.7549,
+      "step": 1830
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00019598765432098764,
+      "loss": 0.732,
+      "step": 1835
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00019567901234567898,
+      "loss": 0.7159,
+      "step": 1840
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00019537037037037035,
+      "loss": 0.7549,
+      "step": 1845
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00019506172839506172,
+      "loss": 0.7581,
+      "step": 1850
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00019475308641975306,
+      "loss": 0.7559,
+      "step": 1855
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00019444444444444443,
+      "loss": 0.7626,
+      "step": 1860
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001941358024691358,
+      "loss": 0.7423,
+      "step": 1865
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019382716049382712,
+      "loss": 0.7672,
+      "step": 1870
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019351851851851849,
+      "loss": 0.7794,
+      "step": 1875
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019320987654320985,
+      "loss": 0.7428,
+      "step": 1880
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019290123456790122,
+      "loss": 0.7462,
+      "step": 1885
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019259259259259257,
+      "loss": 0.702,
+      "step": 1890
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00019228395061728393,
+      "loss": 0.7626,
+      "step": 1895
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001919753086419753,
+      "loss": 0.7357,
+      "step": 1900
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00019166666666666665,
+      "loss": 0.7408,
+      "step": 1905
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00019135802469135801,
+      "loss": 0.707,
+      "step": 1910
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00019104938271604938,
+      "loss": 0.726,
+      "step": 1915
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00019074074074074073,
+      "loss": 0.7396,
+      "step": 1920
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00019043209876543207,
+      "loss": 0.7676,
+      "step": 1925
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00019012345679012344,
+      "loss": 0.7335,
+      "step": 1930
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0001898148148148148,
+      "loss": 0.7235,
+      "step": 1935
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00018950617283950615,
+      "loss": 0.7297,
+      "step": 1940
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018919753086419752,
+      "loss": 0.7079,
+      "step": 1945
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018888888888888888,
+      "loss": 0.7629,
+      "step": 1950
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018858024691358025,
+      "loss": 0.7131,
+      "step": 1955
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018827160493827157,
+      "loss": 0.7517,
+      "step": 1960
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018796296296296294,
+      "loss": 0.7292,
+      "step": 1965
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001876543209876543,
+      "loss": 0.7137,
+      "step": 1970
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00018734567901234565,
+      "loss": 0.725,
+      "step": 1975
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00018703703703703702,
+      "loss": 0.7444,
+      "step": 1980
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00018672839506172839,
+      "loss": 0.7366,
+      "step": 1985
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00018641975308641973,
+      "loss": 0.7329,
+      "step": 1990
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0001861111111111111,
+      "loss": 0.7037,
+      "step": 1995
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00018580246913580247,
+      "loss": 0.7186,
+      "step": 2000
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00018549382716049383,
+      "loss": 0.7322,
+      "step": 2005
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00018518518518518515,
+      "loss": 0.7639,
+      "step": 2010
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00018487654320987652,
+      "loss": 0.7426,
+      "step": 2015
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0001845679012345679,
+      "loss": 0.7648,
+      "step": 2020
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.00018425925925925923,
+      "loss": 0.714,
+      "step": 2025
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0001839506172839506,
+      "loss": 0.7091,
+      "step": 2030
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.00018364197530864197,
+      "loss": 0.7222,
+      "step": 2035
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.00018333333333333334,
+      "loss": 0.7491,
+      "step": 2040
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.00018302469135802468,
+      "loss": 0.7659,
+      "step": 2045
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.00018271604938271605,
+      "loss": 0.727,
+      "step": 2050
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.00018240740740740742,
+      "loss": 0.7416,
+      "step": 2055
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.00018209876543209873,
+      "loss": 0.697,
+      "step": 2060
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.0001817901234567901,
+      "loss": 0.7408,
+      "step": 2065
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00018148148148148147,
+      "loss": 0.7144,
+      "step": 2070
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00018117283950617284,
+      "loss": 0.7745,
+      "step": 2075
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00018086419753086418,
+      "loss": 0.7362,
+      "step": 2080
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00018055555555555555,
+      "loss": 0.7378,
+      "step": 2085
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00018024691358024692,
+      "loss": 0.7227,
+      "step": 2090
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00017993827160493826,
+      "loss": 0.7408,
+      "step": 2095
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.0001796296296296296,
+      "loss": 0.7178,
+      "step": 2100
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00017932098765432097,
+      "loss": 0.7401,
+      "step": 2105
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.0001790123456790123,
+      "loss": 0.7484,
+      "step": 2110
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00017870370370370368,
+      "loss": 0.7271,
+      "step": 2115
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00017839506172839505,
+      "loss": 0.7897,
+      "step": 2120
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00017808641975308642,
+      "loss": 0.7374,
+      "step": 2125
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00017777777777777776,
+      "loss": 0.7314,
+      "step": 2130
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00017746913580246913,
+      "loss": 0.7338,
+      "step": 2135
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.0001771604938271605,
+      "loss": 0.7536,
+      "step": 2140
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.0001768518518518518,
+      "loss": 0.7501,
+      "step": 2145
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00017654320987654318,
+      "loss": 0.7263,
+      "step": 2150
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00017623456790123455,
+      "loss": 0.7468,
+      "step": 2155
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00017592592592592592,
+      "loss": 0.7097,
+      "step": 2160
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00017561728395061726,
+      "loss": 0.7261,
+      "step": 2165
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00017530864197530863,
+      "loss": 0.7489,
+      "step": 2170
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.000175,
+      "loss": 0.7149,
+      "step": 2175
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00017469135802469134,
+      "loss": 0.7239,
+      "step": 2180
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.0001743827160493827,
+      "loss": 0.7249,
+      "step": 2185
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00017407407407407408,
+      "loss": 0.7506,
+      "step": 2190
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0001737654320987654,
+      "loss": 0.7594,
+      "step": 2195
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.00017345679012345676,
+      "loss": 0.7412,
+      "step": 2200
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.00017314814814814813,
+      "loss": 0.7219,
+      "step": 2205
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0001728395061728395,
+      "loss": 0.7655,
+      "step": 2210
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.00017253086419753084,
+      "loss": 0.735,
+      "step": 2215
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.0001722222222222222,
+      "loss": 0.7214,
+      "step": 2220
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00017191358024691358,
+      "loss": 0.7179,
+      "step": 2225
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00017160493827160492,
+      "loss": 0.7208,
+      "step": 2230
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00017129629629629626,
+      "loss": 0.7445,
+      "step": 2235
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00017098765432098763,
+      "loss": 0.7,
+      "step": 2240
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.000170679012345679,
+      "loss": 0.7571,
+      "step": 2245
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.00017037037037037034,
+      "loss": 0.7076,
+      "step": 2250
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0001700617283950617,
+      "loss": 0.7027,
+      "step": 2255
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.00016975308641975308,
+      "loss": 0.7408,
+      "step": 2260
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.00016944444444444442,
+      "loss": 0.7095,
+      "step": 2265
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0001691358024691358,
+      "loss": 0.7314,
+      "step": 2270
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.00016882716049382716,
+      "loss": 0.714,
+      "step": 2275
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.00016851851851851853,
+      "loss": 0.7442,
+      "step": 2280
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.00016820987654320984,
+      "loss": 0.7144,
+      "step": 2285
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0001679012345679012,
+      "loss": 0.7252,
+      "step": 2290
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00016759259259259258,
+      "loss": 0.7118,
+      "step": 2295
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00016728395061728392,
+      "loss": 0.7444,
+      "step": 2300
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.0001669753086419753,
+      "loss": 0.732,
+      "step": 2305
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00016666666666666666,
+      "loss": 0.7394,
+      "step": 2310
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.000166358024691358,
+      "loss": 0.7573,
+      "step": 2315
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.00016604938271604937,
+      "loss": 0.7543,
+      "step": 2320
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.00016574074074074074,
+      "loss": 0.7709,
+      "step": 2325
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0001654320987654321,
+      "loss": 0.7372,
+      "step": 2330
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.00016512345679012342,
+      "loss": 0.7222,
+      "step": 2335
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0001648148148148148,
+      "loss": 0.7505,
+      "step": 2340
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00016450617283950616,
+      "loss": 0.7364,
+      "step": 2345
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.0001641975308641975,
+      "loss": 0.7001,
+      "step": 2350
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00016388888888888887,
+      "loss": 0.7203,
+      "step": 2355
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00016358024691358024,
+      "loss": 0.6906,
+      "step": 2360
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.0001632716049382716,
+      "loss": 0.7557,
+      "step": 2365
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00016296296296296295,
+      "loss": 0.7261,
+      "step": 2370
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.0001626543209876543,
+      "loss": 0.7182,
+      "step": 2375
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00016234567901234566,
+      "loss": 0.7557,
+      "step": 2380
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.000162037037037037,
+      "loss": 0.7251,
+      "step": 2385
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00016172839506172837,
+      "loss": 0.7551,
+      "step": 2390
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00016141975308641974,
+      "loss": 0.6976,
+      "step": 2395
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0001611111111111111,
+      "loss": 0.7336,
+      "step": 2400
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00016080246913580245,
+      "loss": 0.7023,
+      "step": 2405
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00016049382716049382,
+      "loss": 0.7315,
+      "step": 2410
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0001601851851851852,
+      "loss": 0.7092,
+      "step": 2415
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.0001598765432098765,
+      "loss": 0.7381,
+      "step": 2420
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00015956790123456788,
+      "loss": 0.7687,
+      "step": 2425
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00015925925925925924,
+      "loss": 0.7447,
+      "step": 2430
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00015895061728395059,
+      "loss": 0.7476,
+      "step": 2435
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00015864197530864196,
+      "loss": 0.7436,
+      "step": 2440
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00015833333333333332,
+      "loss": 0.7188,
+      "step": 2445
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0001580246913580247,
+      "loss": 0.7284,
+      "step": 2450
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00015771604938271603,
+      "loss": 0.739,
+      "step": 2455
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0001574074074074074,
+      "loss": 0.7108,
+      "step": 2460
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00015709876543209877,
+      "loss": 0.7101,
+      "step": 2465
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0001567901234567901,
+      "loss": 0.7433,
+      "step": 2470
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.00015648148148148146,
+      "loss": 0.7009,
+      "step": 2475
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.00015617283950617283,
+      "loss": 0.7046,
+      "step": 2480
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0001558641975308642,
+      "loss": 0.7267,
+      "step": 2485
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.00015555555555555554,
+      "loss": 0.7253,
+      "step": 2490
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.0001552469135802469,
+      "loss": 0.6932,
+      "step": 2495
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.00015493827160493827,
+      "loss": 0.7542,
+      "step": 2500
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 5010,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 500,
+  "total_flos": 1.006726589213311e+19,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

save_lora_instruct_r8_3e4_s11_1/checkpoint-2500/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6bc4247ac8c6f92ef49080d32b2520dd809daff76a4805fdafb6e454ee1c5cc3
+size 5944

save_lora_instruct_r8_3e4_s11_1/checkpoint-2500/zero_to_fp32.py

@@ -0,0 +1,587 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel()
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)

save_lora_instruct_r8_3e4_s11_1/checkpoint-3000/README.md

@@ -0,0 +1,202 @@
+---
+base_model: meta-llama/Llama-2-7b-hf
+library_name: peft
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.9.0

save_lora_instruct_r8_3e4_s11_1/checkpoint-3000/adapter_config.json

@@ -0,0 +1,35 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "meta-llama/Llama-2-7b-hf",
+  "bias": "none",
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 16,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "mora_type": 1,
+  "peft_type": "LORA",
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "k_proj",
+    "v_proj",
+    "o_proj",
+    "gate_proj",
+    "down_proj",
+    "up_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_mora": false,
+  "use_rslora": false
+}

save_lora_instruct_r8_3e4_s11_1/checkpoint-3000/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f5f218954c06132c08fa28fca5c2433ca64ab44ea4ad93dcdb5a731c3ef1540
+size 564342240

save_lora_instruct_r8_3e4_s11_1/checkpoint-3000/global_step3000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a4955e83581d97509825145193cd93905e5781d8087d55be39f378ddd0517ca1
+size 239914224

save_lora_instruct_r8_3e4_s11_1/checkpoint-3000/global_step3000/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec57ac2158f753326591fe3943d0879036cd7ab5b43c5558d416434ec9080229
+size 40303404

save_lora_instruct_r8_3e4_s11_1/checkpoint-3000/latest

@@ -0,0 +1 @@
+global_step3000

save_lora_instruct_r8_3e4_s11_1/checkpoint-3000/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:602ff114cd2710380875f9cf15b09f6fc6a211d90bdfc8416e511f12377a1bd9
+size 14244

save_lora_instruct_r8_3e4_s11_1/checkpoint-3000/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61a768c931ed1c4d8a2f34684c8f3d70a05baa14bed2ec907bcb93ae3650dbaf
+size 1064

save_lora_instruct_r8_3e4_s11_1/checkpoint-3000/trainer_state.json

@@ -0,0 +1,3621 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.197492749555618,
+  "eval_steps": 500,
+  "global_step": 3000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.999999999999999e-06,
+      "loss": 1.38,
+      "step": 5
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 1.9999999999999998e-05,
+      "loss": 1.3936,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 2.9999999999999997e-05,
+      "loss": 1.3006,
+      "step": 15
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.9999999999999996e-05,
+      "loss": 1.2594,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 4.9999999999999996e-05,
+      "loss": 1.1177,
+      "step": 25
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 5.9999999999999995e-05,
+      "loss": 1.0481,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 7e-05,
+      "loss": 1.0601,
+      "step": 35
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 7.999999999999999e-05,
+      "loss": 1.0473,
+      "step": 40
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 8.999999999999999e-05,
+      "loss": 0.9523,
+      "step": 45
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.999999999999999e-05,
+      "loss": 0.9451,
+      "step": 50
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00010999999999999998,
+      "loss": 0.9687,
+      "step": 55
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00011999999999999999,
+      "loss": 0.9054,
+      "step": 60
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00013,
+      "loss": 0.9518,
+      "step": 65
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00014,
+      "loss": 0.8763,
+      "step": 70
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00015,
+      "loss": 0.8657,
+      "step": 75
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00015999999999999999,
+      "loss": 0.9145,
+      "step": 80
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00016999999999999999,
+      "loss": 0.8296,
+      "step": 85
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00017999999999999998,
+      "loss": 0.8709,
+      "step": 90
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00018999999999999998,
+      "loss": 0.8405,
+      "step": 95
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00019999999999999998,
+      "loss": 0.8604,
+      "step": 100
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00020999999999999998,
+      "loss": 0.8607,
+      "step": 105
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00021999999999999995,
+      "loss": 0.8376,
+      "step": 110
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00023,
+      "loss": 0.8554,
+      "step": 115
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00023999999999999998,
+      "loss": 0.8456,
+      "step": 120
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00025,
+      "loss": 0.8443,
+      "step": 125
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00026,
+      "loss": 0.8327,
+      "step": 130
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00027,
+      "loss": 0.8096,
+      "step": 135
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00028,
+      "loss": 0.7972,
+      "step": 140
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00029,
+      "loss": 0.8306,
+      "step": 145
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0003,
+      "loss": 0.8409,
+      "step": 150
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0002996913580246913,
+      "loss": 0.8094,
+      "step": 155
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00029938271604938266,
+      "loss": 0.8146,
+      "step": 160
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00029907407407407405,
+      "loss": 0.8042,
+      "step": 165
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0002987654320987654,
+      "loss": 0.7784,
+      "step": 170
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0002984567901234568,
+      "loss": 0.828,
+      "step": 175
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00029814814814814813,
+      "loss": 0.8473,
+      "step": 180
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0002978395061728395,
+      "loss": 0.8354,
+      "step": 185
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00029753086419753087,
+      "loss": 0.8246,
+      "step": 190
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00029722222222222216,
+      "loss": 0.8249,
+      "step": 195
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00029691358024691355,
+      "loss": 0.8222,
+      "step": 200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0002966049382716049,
+      "loss": 0.8369,
+      "step": 205
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00029629629629629624,
+      "loss": 0.7954,
+      "step": 210
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00029598765432098763,
+      "loss": 0.8051,
+      "step": 215
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000295679012345679,
+      "loss": 0.8081,
+      "step": 220
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00029537037037037037,
+      "loss": 0.7864,
+      "step": 225
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0002950617283950617,
+      "loss": 0.7974,
+      "step": 230
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00029475308641975306,
+      "loss": 0.8306,
+      "step": 235
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00029444444444444445,
+      "loss": 0.7792,
+      "step": 240
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00029413580246913574,
+      "loss": 0.8108,
+      "step": 245
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00029382716049382714,
+      "loss": 0.8167,
+      "step": 250
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0002935185185185185,
+      "loss": 0.8358,
+      "step": 255
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0002932098765432099,
+      "loss": 0.7948,
+      "step": 260
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0002929012345679012,
+      "loss": 0.8078,
+      "step": 265
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00029259259259259256,
+      "loss": 0.7886,
+      "step": 270
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00029228395061728395,
+      "loss": 0.7903,
+      "step": 275
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0002919753086419753,
+      "loss": 0.7892,
+      "step": 280
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00029166666666666664,
+      "loss": 0.7717,
+      "step": 285
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.000291358024691358,
+      "loss": 0.7774,
+      "step": 290
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0002910493827160494,
+      "loss": 0.8053,
+      "step": 295
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0002907407407407407,
+      "loss": 0.794,
+      "step": 300
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00029043209876543206,
+      "loss": 0.7856,
+      "step": 305
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00029012345679012345,
+      "loss": 0.7713,
+      "step": 310
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0002898148148148148,
+      "loss": 0.7805,
+      "step": 315
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00028950617283950614,
+      "loss": 0.7897,
+      "step": 320
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00028919753086419753,
+      "loss": 0.7931,
+      "step": 325
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0002888888888888888,
+      "loss": 0.7739,
+      "step": 330
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0002885802469135802,
+      "loss": 0.7749,
+      "step": 335
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00028827160493827156,
+      "loss": 0.7932,
+      "step": 340
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00028796296296296296,
+      "loss": 0.7906,
+      "step": 345
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0002876543209876543,
+      "loss": 0.8094,
+      "step": 350
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00028734567901234564,
+      "loss": 0.7932,
+      "step": 355
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00028703703703703703,
+      "loss": 0.8027,
+      "step": 360
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0002867283950617284,
+      "loss": 0.8297,
+      "step": 365
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0002864197530864197,
+      "loss": 0.8228,
+      "step": 370
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0002861111111111111,
+      "loss": 0.7997,
+      "step": 375
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00028580246913580246,
+      "loss": 0.8124,
+      "step": 380
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0002854938271604938,
+      "loss": 0.8152,
+      "step": 385
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00028518518518518514,
+      "loss": 0.7833,
+      "step": 390
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00028487654320987654,
+      "loss": 0.7884,
+      "step": 395
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0002845679012345679,
+      "loss": 0.753,
+      "step": 400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0002842592592592592,
+      "loss": 0.8055,
+      "step": 405
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0002839506172839506,
+      "loss": 0.8014,
+      "step": 410
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00028364197530864196,
+      "loss": 0.782,
+      "step": 415
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0002833333333333333,
+      "loss": 0.7733,
+      "step": 420
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0002830246913580247,
+      "loss": 0.7777,
+      "step": 425
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00028271604938271604,
+      "loss": 0.752,
+      "step": 430
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0002824074074074074,
+      "loss": 0.7963,
+      "step": 435
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0002820987654320987,
+      "loss": 0.7655,
+      "step": 440
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0002817901234567901,
+      "loss": 0.8,
+      "step": 445
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00028148148148148146,
+      "loss": 0.7465,
+      "step": 450
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0002811728395061728,
+      "loss": 0.7432,
+      "step": 455
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0002808641975308642,
+      "loss": 0.758,
+      "step": 460
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00028055555555555554,
+      "loss": 0.7757,
+      "step": 465
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0002802469135802469,
+      "loss": 0.7816,
+      "step": 470
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0002799382716049382,
+      "loss": 0.7744,
+      "step": 475
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0002796296296296296,
+      "loss": 0.7807,
+      "step": 480
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00027932098765432096,
+      "loss": 0.774,
+      "step": 485
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0002790123456790123,
+      "loss": 0.7361,
+      "step": 490
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0002787037037037037,
+      "loss": 0.7663,
+      "step": 495
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00027839506172839504,
+      "loss": 0.7532,
+      "step": 500
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0002780864197530864,
+      "loss": 0.7573,
+      "step": 505
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0002777777777777778,
+      "loss": 0.7802,
+      "step": 510
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0002774691358024691,
+      "loss": 0.7871,
+      "step": 515
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00027716049382716046,
+      "loss": 0.7734,
+      "step": 520
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0002768518518518518,
+      "loss": 0.7752,
+      "step": 525
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0002765432098765432,
+      "loss": 0.7697,
+      "step": 530
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00027623456790123454,
+      "loss": 0.7846,
+      "step": 535
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0002759259259259259,
+      "loss": 0.7592,
+      "step": 540
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0002756172839506173,
+      "loss": 0.7734,
+      "step": 545
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0002753086419753086,
+      "loss": 0.7984,
+      "step": 550
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00027499999999999996,
+      "loss": 0.7342,
+      "step": 555
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00027469135802469136,
+      "loss": 0.7479,
+      "step": 560
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0002743827160493827,
+      "loss": 0.819,
+      "step": 565
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00027407407407407404,
+      "loss": 0.7561,
+      "step": 570
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0002737654320987654,
+      "loss": 0.7303,
+      "step": 575
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0002734567901234568,
+      "loss": 0.8001,
+      "step": 580
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0002731481481481481,
+      "loss": 0.7738,
+      "step": 585
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00027283950617283946,
+      "loss": 0.7866,
+      "step": 590
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00027253086419753086,
+      "loss": 0.7582,
+      "step": 595
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0002722222222222222,
+      "loss": 0.7645,
+      "step": 600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00027191358024691354,
+      "loss": 0.7839,
+      "step": 605
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0002716049382716049,
+      "loss": 0.7931,
+      "step": 610
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0002712962962962963,
+      "loss": 0.7692,
+      "step": 615
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0002709876543209876,
+      "loss": 0.7835,
+      "step": 620
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00027067901234567896,
+      "loss": 0.7795,
+      "step": 625
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00027037037037037036,
+      "loss": 0.7639,
+      "step": 630
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0002700617283950617,
+      "loss": 0.7705,
+      "step": 635
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00026975308641975304,
+      "loss": 0.7674,
+      "step": 640
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00026944444444444444,
+      "loss": 0.7349,
+      "step": 645
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0002691358024691358,
+      "loss": 0.8008,
+      "step": 650
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0002688271604938271,
+      "loss": 0.755,
+      "step": 655
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00026851851851851847,
+      "loss": 0.7526,
+      "step": 660
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00026820987654320986,
+      "loss": 0.7611,
+      "step": 665
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0002679012345679012,
+      "loss": 0.7621,
+      "step": 670
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00026759259259259255,
+      "loss": 0.7768,
+      "step": 675
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00026728395061728394,
+      "loss": 0.7703,
+      "step": 680
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0002669753086419753,
+      "loss": 0.7667,
+      "step": 685
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0002666666666666666,
+      "loss": 0.7712,
+      "step": 690
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.000266358024691358,
+      "loss": 0.7269,
+      "step": 695
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00026604938271604936,
+      "loss": 0.7275,
+      "step": 700
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0002657407407407407,
+      "loss": 0.7786,
+      "step": 705
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00026543209876543205,
+      "loss": 0.7808,
+      "step": 710
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00026512345679012344,
+      "loss": 0.7455,
+      "step": 715
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0002648148148148148,
+      "loss": 0.7717,
+      "step": 720
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0002645061728395061,
+      "loss": 0.737,
+      "step": 725
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0002641975308641975,
+      "loss": 0.7664,
+      "step": 730
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00026388888888888886,
+      "loss": 0.7716,
+      "step": 735
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00026358024691358026,
+      "loss": 0.7638,
+      "step": 740
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00026327160493827155,
+      "loss": 0.7777,
+      "step": 745
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00026296296296296294,
+      "loss": 0.7431,
+      "step": 750
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0002626543209876543,
+      "loss": 0.7516,
+      "step": 755
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00026234567901234563,
+      "loss": 0.831,
+      "step": 760
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.000262037037037037,
+      "loss": 0.7423,
+      "step": 765
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00026172839506172837,
+      "loss": 0.7471,
+      "step": 770
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0002614197530864197,
+      "loss": 0.7551,
+      "step": 775
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0002611111111111111,
+      "loss": 0.7208,
+      "step": 780
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00026080246913580245,
+      "loss": 0.7359,
+      "step": 785
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00026049382716049384,
+      "loss": 0.7759,
+      "step": 790
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00026018518518518513,
+      "loss": 0.7406,
+      "step": 795
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0002598765432098765,
+      "loss": 0.756,
+      "step": 800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00025956790123456787,
+      "loss": 0.7631,
+      "step": 805
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0002592592592592592,
+      "loss": 0.7885,
+      "step": 810
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0002589506172839506,
+      "loss": 0.7947,
+      "step": 815
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00025864197530864195,
+      "loss": 0.7557,
+      "step": 820
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00025833333333333334,
+      "loss": 0.7556,
+      "step": 825
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0002580246913580247,
+      "loss": 0.7472,
+      "step": 830
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.000257716049382716,
+      "loss": 0.729,
+      "step": 835
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0002574074074074074,
+      "loss": 0.7523,
+      "step": 840
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0002570987654320987,
+      "loss": 0.7505,
+      "step": 845
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0002567901234567901,
+      "loss": 0.7617,
+      "step": 850
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00025648148148148145,
+      "loss": 0.7898,
+      "step": 855
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00025617283950617284,
+      "loss": 0.7575,
+      "step": 860
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0002558641975308642,
+      "loss": 0.8316,
+      "step": 865
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00025555555555555553,
+      "loss": 0.7261,
+      "step": 870
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0002552469135802469,
+      "loss": 0.7594,
+      "step": 875
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00025493827160493826,
+      "loss": 0.7372,
+      "step": 880
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0002546296296296296,
+      "loss": 0.7498,
+      "step": 885
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00025432098765432095,
+      "loss": 0.7398,
+      "step": 890
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0002540123456790123,
+      "loss": 0.804,
+      "step": 895
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0002537037037037037,
+      "loss": 0.7241,
+      "step": 900
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00025339506172839503,
+      "loss": 0.7519,
+      "step": 905
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0002530864197530864,
+      "loss": 0.7694,
+      "step": 910
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00025277777777777777,
+      "loss": 0.7474,
+      "step": 915
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0002524691358024691,
+      "loss": 0.7278,
+      "step": 920
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0002521604938271605,
+      "loss": 0.7747,
+      "step": 925
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0002518518518518518,
+      "loss": 0.7858,
+      "step": 930
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0002515432098765432,
+      "loss": 0.7442,
+      "step": 935
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00025123456790123453,
+      "loss": 0.7446,
+      "step": 940
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0002509259259259259,
+      "loss": 0.7843,
+      "step": 945
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00025061728395061727,
+      "loss": 0.7512,
+      "step": 950
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0002503086419753086,
+      "loss": 0.7616,
+      "step": 955
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00025,
+      "loss": 0.826,
+      "step": 960
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00024969135802469135,
+      "loss": 0.7567,
+      "step": 965
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0002493827160493827,
+      "loss": 0.7754,
+      "step": 970
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0002490740740740741,
+      "loss": 0.7605,
+      "step": 975
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00024876543209876537,
+      "loss": 0.7728,
+      "step": 980
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00024845679012345677,
+      "loss": 0.7902,
+      "step": 985
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0002481481481481481,
+      "loss": 0.7622,
+      "step": 990
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0002478395061728395,
+      "loss": 0.745,
+      "step": 995
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00024753086419753085,
+      "loss": 0.7529,
+      "step": 1000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0002472222222222222,
+      "loss": 0.759,
+      "step": 1005
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0002469135802469136,
+      "loss": 0.7471,
+      "step": 1010
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00024660493827160493,
+      "loss": 0.7517,
+      "step": 1015
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00024629629629629627,
+      "loss": 0.7634,
+      "step": 1020
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002459876543209876,
+      "loss": 0.7833,
+      "step": 1025
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.000245679012345679,
+      "loss": 0.7227,
+      "step": 1030
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00024537037037037035,
+      "loss": 0.7792,
+      "step": 1035
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002450617283950617,
+      "loss": 0.7292,
+      "step": 1040
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002447530864197531,
+      "loss": 0.7519,
+      "step": 1045
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00024444444444444443,
+      "loss": 0.7385,
+      "step": 1050
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00024413580246913577,
+      "loss": 0.7842,
+      "step": 1055
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00024382716049382714,
+      "loss": 0.796,
+      "step": 1060
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002435185185185185,
+      "loss": 0.7479,
+      "step": 1065
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00024320987654320985,
+      "loss": 0.7531,
+      "step": 1070
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00024290123456790122,
+      "loss": 0.731,
+      "step": 1075
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002425925925925926,
+      "loss": 0.7319,
+      "step": 1080
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00024228395061728393,
+      "loss": 0.7381,
+      "step": 1085
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024197530864197527,
+      "loss": 0.7621,
+      "step": 1090
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024166666666666664,
+      "loss": 0.771,
+      "step": 1095
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024135802469135798,
+      "loss": 0.7493,
+      "step": 1100
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024104938271604935,
+      "loss": 0.7553,
+      "step": 1105
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024074074074074072,
+      "loss": 0.7862,
+      "step": 1110
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002404320987654321,
+      "loss": 0.7531,
+      "step": 1115
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00024012345679012343,
+      "loss": 0.7688,
+      "step": 1120
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002398148148148148,
+      "loss": 0.7666,
+      "step": 1125
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00023950617283950617,
+      "loss": 0.7484,
+      "step": 1130
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00023919753086419748,
+      "loss": 0.8048,
+      "step": 1135
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00023888888888888885,
+      "loss": 0.757,
+      "step": 1140
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00023858024691358022,
+      "loss": 0.7838,
+      "step": 1145
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002382716049382716,
+      "loss": 0.7834,
+      "step": 1150
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00023796296296296293,
+      "loss": 0.733,
+      "step": 1155
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002376543209876543,
+      "loss": 0.7195,
+      "step": 1160
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00023734567901234567,
+      "loss": 0.7478,
+      "step": 1165
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.000237037037037037,
+      "loss": 0.7456,
+      "step": 1170
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00023672839506172838,
+      "loss": 0.722,
+      "step": 1175
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00023641975308641975,
+      "loss": 0.7395,
+      "step": 1180
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00023611111111111112,
+      "loss": 0.7553,
+      "step": 1185
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00023580246913580243,
+      "loss": 0.7692,
+      "step": 1190
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002354938271604938,
+      "loss": 0.7396,
+      "step": 1195
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00023518518518518517,
+      "loss": 0.7214,
+      "step": 1200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002348765432098765,
+      "loss": 0.7359,
+      "step": 1205
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00023456790123456788,
+      "loss": 0.7666,
+      "step": 1210
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00023425925925925925,
+      "loss": 0.7583,
+      "step": 1215
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002339506172839506,
+      "loss": 0.7715,
+      "step": 1220
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00023364197530864196,
+      "loss": 0.7423,
+      "step": 1225
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002333333333333333,
+      "loss": 0.7598,
+      "step": 1230
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00023302469135802467,
+      "loss": 0.7251,
+      "step": 1235
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00023271604938271601,
+      "loss": 0.7447,
+      "step": 1240
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00023240740740740738,
+      "loss": 0.7923,
+      "step": 1245
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00023209876543209875,
+      "loss": 0.7161,
+      "step": 1250
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002317901234567901,
+      "loss": 0.7291,
+      "step": 1255
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00023148148148148146,
+      "loss": 0.7366,
+      "step": 1260
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00023117283950617283,
+      "loss": 0.7173,
+      "step": 1265
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002308641975308642,
+      "loss": 0.7575,
+      "step": 1270
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00023055555555555552,
+      "loss": 0.7632,
+      "step": 1275
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00023024691358024688,
+      "loss": 0.7601,
+      "step": 1280
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00022993827160493825,
+      "loss": 0.7395,
+      "step": 1285
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002296296296296296,
+      "loss": 0.7686,
+      "step": 1290
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00022932098765432096,
+      "loss": 0.7531,
+      "step": 1295
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00022901234567901233,
+      "loss": 0.7675,
+      "step": 1300
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002287037037037037,
+      "loss": 0.8103,
+      "step": 1305
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00022839506172839504,
+      "loss": 0.773,
+      "step": 1310
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002280864197530864,
+      "loss": 0.7549,
+      "step": 1315
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00022777777777777778,
+      "loss": 0.7547,
+      "step": 1320
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002274691358024691,
+      "loss": 0.758,
+      "step": 1325
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00022716049382716047,
+      "loss": 0.77,
+      "step": 1330
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00022685185185185183,
+      "loss": 0.7662,
+      "step": 1335
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00022654320987654318,
+      "loss": 0.744,
+      "step": 1340
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022623456790123455,
+      "loss": 0.7486,
+      "step": 1345
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022592592592592591,
+      "loss": 0.7473,
+      "step": 1350
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022561728395061728,
+      "loss": 0.7486,
+      "step": 1355
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022530864197530862,
+      "loss": 0.7205,
+      "step": 1360
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.000225,
+      "loss": 0.7289,
+      "step": 1365
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022469135802469134,
+      "loss": 0.7586,
+      "step": 1370
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022438271604938268,
+      "loss": 0.7175,
+      "step": 1375
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022407407407407405,
+      "loss": 0.6961,
+      "step": 1380
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022376543209876542,
+      "loss": 0.7242,
+      "step": 1385
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022345679012345678,
+      "loss": 0.7515,
+      "step": 1390
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022314814814814813,
+      "loss": 0.7522,
+      "step": 1395
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002228395061728395,
+      "loss": 0.7651,
+      "step": 1400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022253086419753086,
+      "loss": 0.6958,
+      "step": 1405
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022222222222222218,
+      "loss": 0.7336,
+      "step": 1410
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022191358024691355,
+      "loss": 0.7636,
+      "step": 1415
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00022160493827160492,
+      "loss": 0.7583,
+      "step": 1420
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00022129629629629626,
+      "loss": 0.7605,
+      "step": 1425
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00022098765432098763,
+      "loss": 0.7109,
+      "step": 1430
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.000220679012345679,
+      "loss": 0.7224,
+      "step": 1435
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00022037037037037037,
+      "loss": 0.7724,
+      "step": 1440
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002200617283950617,
+      "loss": 0.7637,
+      "step": 1445
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021975308641975308,
+      "loss": 0.7302,
+      "step": 1450
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021944444444444444,
+      "loss": 0.7685,
+      "step": 1455
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021913580246913576,
+      "loss": 0.751,
+      "step": 1460
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021882716049382713,
+      "loss": 0.7768,
+      "step": 1465
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002185185185185185,
+      "loss": 0.7461,
+      "step": 1470
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00021820987654320987,
+      "loss": 0.7782,
+      "step": 1475
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002179012345679012,
+      "loss": 0.7474,
+      "step": 1480
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00021759259259259258,
+      "loss": 0.7475,
+      "step": 1485
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00021728395061728395,
+      "loss": 0.7539,
+      "step": 1490
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002169753086419753,
+      "loss": 0.7476,
+      "step": 1495
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00021666666666666666,
+      "loss": 0.7382,
+      "step": 1500
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.000216358024691358,
+      "loss": 0.7305,
+      "step": 1505
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00021604938271604937,
+      "loss": 0.7455,
+      "step": 1510
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002157407407407407,
+      "loss": 0.7286,
+      "step": 1515
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00021543209876543208,
+      "loss": 0.7653,
+      "step": 1520
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00021512345679012345,
+      "loss": 0.7443,
+      "step": 1525
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0002148148148148148,
+      "loss": 0.7026,
+      "step": 1530
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00021450617283950616,
+      "loss": 0.7436,
+      "step": 1535
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00021419753086419753,
+      "loss": 0.7101,
+      "step": 1540
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00021388888888888884,
+      "loss": 0.7484,
+      "step": 1545
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002135802469135802,
+      "loss": 0.7483,
+      "step": 1550
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00021327160493827158,
+      "loss": 0.7419,
+      "step": 1555
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00021296296296296295,
+      "loss": 0.7356,
+      "step": 1560
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002126543209876543,
+      "loss": 0.7716,
+      "step": 1565
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00021234567901234566,
+      "loss": 0.734,
+      "step": 1570
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00021203703703703703,
+      "loss": 0.7268,
+      "step": 1575
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00021172839506172837,
+      "loss": 0.7133,
+      "step": 1580
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00021141975308641974,
+      "loss": 0.7575,
+      "step": 1585
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002111111111111111,
+      "loss": 0.7219,
+      "step": 1590
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00021080246913580248,
+      "loss": 0.7394,
+      "step": 1595
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0002104938271604938,
+      "loss": 0.7729,
+      "step": 1600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00021018518518518516,
+      "loss": 0.757,
+      "step": 1605
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00020987654320987653,
+      "loss": 0.7139,
+      "step": 1610
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00020956790123456787,
+      "loss": 0.7492,
+      "step": 1615
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00020925925925925924,
+      "loss": 0.7461,
+      "step": 1620
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002089506172839506,
+      "loss": 0.7276,
+      "step": 1625
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00020864197530864198,
+      "loss": 0.7725,
+      "step": 1630
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00020833333333333332,
+      "loss": 0.7623,
+      "step": 1635
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002080246913580247,
+      "loss": 0.7394,
+      "step": 1640
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00020771604938271603,
+      "loss": 0.7387,
+      "step": 1645
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00020740740740740737,
+      "loss": 0.7519,
+      "step": 1650
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00020709876543209874,
+      "loss": 0.7625,
+      "step": 1655
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002067901234567901,
+      "loss": 0.7254,
+      "step": 1660
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00020648148148148145,
+      "loss": 0.7129,
+      "step": 1665
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00020617283950617282,
+      "loss": 0.7964,
+      "step": 1670
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002058641975308642,
+      "loss": 0.7332,
+      "step": 1675
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00020555555555555556,
+      "loss": 0.7955,
+      "step": 1680
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00020524691358024687,
+      "loss": 0.7544,
+      "step": 1685
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00020493827160493824,
+      "loss": 0.7115,
+      "step": 1690
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002046296296296296,
+      "loss": 0.7434,
+      "step": 1695
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00020432098765432095,
+      "loss": 0.7382,
+      "step": 1700
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00020401234567901232,
+      "loss": 0.7452,
+      "step": 1705
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002037037037037037,
+      "loss": 0.7551,
+      "step": 1710
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00020339506172839506,
+      "loss": 0.7202,
+      "step": 1715
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002030864197530864,
+      "loss": 0.735,
+      "step": 1720
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00020277777777777777,
+      "loss": 0.7581,
+      "step": 1725
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00020246913580246914,
+      "loss": 0.7527,
+      "step": 1730
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00020216049382716045,
+      "loss": 0.7498,
+      "step": 1735
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00020185185185185182,
+      "loss": 0.735,
+      "step": 1740
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002015432098765432,
+      "loss": 0.7286,
+      "step": 1745
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00020123456790123456,
+      "loss": 0.6891,
+      "step": 1750
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002009259259259259,
+      "loss": 0.7353,
+      "step": 1755
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00020061728395061727,
+      "loss": 0.7386,
+      "step": 1760
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00020030864197530864,
+      "loss": 0.7106,
+      "step": 1765
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00019999999999999998,
+      "loss": 0.7207,
+      "step": 1770
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00019969135802469135,
+      "loss": 0.7576,
+      "step": 1775
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00019938271604938272,
+      "loss": 0.7462,
+      "step": 1780
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00019907407407407404,
+      "loss": 0.7068,
+      "step": 1785
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001987654320987654,
+      "loss": 0.7525,
+      "step": 1790
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019845679012345677,
+      "loss": 0.7291,
+      "step": 1795
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019814814814814814,
+      "loss": 0.7226,
+      "step": 1800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019783950617283948,
+      "loss": 0.7445,
+      "step": 1805
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019753086419753085,
+      "loss": 0.7571,
+      "step": 1810
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019722222222222222,
+      "loss": 0.7306,
+      "step": 1815
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00019691358024691354,
+      "loss": 0.7785,
+      "step": 1820
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001966049382716049,
+      "loss": 0.7537,
+      "step": 1825
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00019629629629629627,
+      "loss": 0.7549,
+      "step": 1830
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00019598765432098764,
+      "loss": 0.732,
+      "step": 1835
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00019567901234567898,
+      "loss": 0.7159,
+      "step": 1840
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00019537037037037035,
+      "loss": 0.7549,
+      "step": 1845
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00019506172839506172,
+      "loss": 0.7581,
+      "step": 1850
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00019475308641975306,
+      "loss": 0.7559,
+      "step": 1855
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00019444444444444443,
+      "loss": 0.7626,
+      "step": 1860
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001941358024691358,
+      "loss": 0.7423,
+      "step": 1865
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019382716049382712,
+      "loss": 0.7672,
+      "step": 1870
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019351851851851849,
+      "loss": 0.7794,
+      "step": 1875
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019320987654320985,
+      "loss": 0.7428,
+      "step": 1880
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019290123456790122,
+      "loss": 0.7462,
+      "step": 1885
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019259259259259257,
+      "loss": 0.702,
+      "step": 1890
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00019228395061728393,
+      "loss": 0.7626,
+      "step": 1895
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001919753086419753,
+      "loss": 0.7357,
+      "step": 1900
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00019166666666666665,
+      "loss": 0.7408,
+      "step": 1905
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00019135802469135801,
+      "loss": 0.707,
+      "step": 1910
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00019104938271604938,
+      "loss": 0.726,
+      "step": 1915
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00019074074074074073,
+      "loss": 0.7396,
+      "step": 1920
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00019043209876543207,
+      "loss": 0.7676,
+      "step": 1925
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00019012345679012344,
+      "loss": 0.7335,
+      "step": 1930
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0001898148148148148,
+      "loss": 0.7235,
+      "step": 1935
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00018950617283950615,
+      "loss": 0.7297,
+      "step": 1940
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018919753086419752,
+      "loss": 0.7079,
+      "step": 1945
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018888888888888888,
+      "loss": 0.7629,
+      "step": 1950
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018858024691358025,
+      "loss": 0.7131,
+      "step": 1955
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018827160493827157,
+      "loss": 0.7517,
+      "step": 1960
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018796296296296294,
+      "loss": 0.7292,
+      "step": 1965
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001876543209876543,
+      "loss": 0.7137,
+      "step": 1970
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00018734567901234565,
+      "loss": 0.725,
+      "step": 1975
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00018703703703703702,
+      "loss": 0.7444,
+      "step": 1980
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00018672839506172839,
+      "loss": 0.7366,
+      "step": 1985
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00018641975308641973,
+      "loss": 0.7329,
+      "step": 1990
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0001861111111111111,
+      "loss": 0.7037,
+      "step": 1995
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00018580246913580247,
+      "loss": 0.7186,
+      "step": 2000
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00018549382716049383,
+      "loss": 0.7322,
+      "step": 2005
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00018518518518518515,
+      "loss": 0.7639,
+      "step": 2010
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00018487654320987652,
+      "loss": 0.7426,
+      "step": 2015
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0001845679012345679,
+      "loss": 0.7648,
+      "step": 2020
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.00018425925925925923,
+      "loss": 0.714,
+      "step": 2025
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0001839506172839506,
+      "loss": 0.7091,
+      "step": 2030
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.00018364197530864197,
+      "loss": 0.7222,
+      "step": 2035
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.00018333333333333334,
+      "loss": 0.7491,
+      "step": 2040
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.00018302469135802468,
+      "loss": 0.7659,
+      "step": 2045
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.00018271604938271605,
+      "loss": 0.727,
+      "step": 2050
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.00018240740740740742,
+      "loss": 0.7416,
+      "step": 2055
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.00018209876543209873,
+      "loss": 0.697,
+      "step": 2060
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.0001817901234567901,
+      "loss": 0.7408,
+      "step": 2065
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00018148148148148147,
+      "loss": 0.7144,
+      "step": 2070
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00018117283950617284,
+      "loss": 0.7745,
+      "step": 2075
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00018086419753086418,
+      "loss": 0.7362,
+      "step": 2080
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00018055555555555555,
+      "loss": 0.7378,
+      "step": 2085
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00018024691358024692,
+      "loss": 0.7227,
+      "step": 2090
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00017993827160493826,
+      "loss": 0.7408,
+      "step": 2095
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.0001796296296296296,
+      "loss": 0.7178,
+      "step": 2100
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00017932098765432097,
+      "loss": 0.7401,
+      "step": 2105
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.0001790123456790123,
+      "loss": 0.7484,
+      "step": 2110
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00017870370370370368,
+      "loss": 0.7271,
+      "step": 2115
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00017839506172839505,
+      "loss": 0.7897,
+      "step": 2120
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00017808641975308642,
+      "loss": 0.7374,
+      "step": 2125
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00017777777777777776,
+      "loss": 0.7314,
+      "step": 2130
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00017746913580246913,
+      "loss": 0.7338,
+      "step": 2135
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.0001771604938271605,
+      "loss": 0.7536,
+      "step": 2140
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.0001768518518518518,
+      "loss": 0.7501,
+      "step": 2145
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00017654320987654318,
+      "loss": 0.7263,
+      "step": 2150
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00017623456790123455,
+      "loss": 0.7468,
+      "step": 2155
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00017592592592592592,
+      "loss": 0.7097,
+      "step": 2160
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00017561728395061726,
+      "loss": 0.7261,
+      "step": 2165
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00017530864197530863,
+      "loss": 0.7489,
+      "step": 2170
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.000175,
+      "loss": 0.7149,
+      "step": 2175
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00017469135802469134,
+      "loss": 0.7239,
+      "step": 2180
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.0001743827160493827,
+      "loss": 0.7249,
+      "step": 2185
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00017407407407407408,
+      "loss": 0.7506,
+      "step": 2190
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0001737654320987654,
+      "loss": 0.7594,
+      "step": 2195
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.00017345679012345676,
+      "loss": 0.7412,
+      "step": 2200
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.00017314814814814813,
+      "loss": 0.7219,
+      "step": 2205
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0001728395061728395,
+      "loss": 0.7655,
+      "step": 2210
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.00017253086419753084,
+      "loss": 0.735,
+      "step": 2215
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.0001722222222222222,
+      "loss": 0.7214,
+      "step": 2220
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00017191358024691358,
+      "loss": 0.7179,
+      "step": 2225
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00017160493827160492,
+      "loss": 0.7208,
+      "step": 2230
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00017129629629629626,
+      "loss": 0.7445,
+      "step": 2235
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00017098765432098763,
+      "loss": 0.7,
+      "step": 2240
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.000170679012345679,
+      "loss": 0.7571,
+      "step": 2245
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.00017037037037037034,
+      "loss": 0.7076,
+      "step": 2250
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0001700617283950617,
+      "loss": 0.7027,
+      "step": 2255
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.00016975308641975308,
+      "loss": 0.7408,
+      "step": 2260
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.00016944444444444442,
+      "loss": 0.7095,
+      "step": 2265
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0001691358024691358,
+      "loss": 0.7314,
+      "step": 2270
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.00016882716049382716,
+      "loss": 0.714,
+      "step": 2275
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.00016851851851851853,
+      "loss": 0.7442,
+      "step": 2280
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.00016820987654320984,
+      "loss": 0.7144,
+      "step": 2285
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0001679012345679012,
+      "loss": 0.7252,
+      "step": 2290
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00016759259259259258,
+      "loss": 0.7118,
+      "step": 2295
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00016728395061728392,
+      "loss": 0.7444,
+      "step": 2300
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.0001669753086419753,
+      "loss": 0.732,
+      "step": 2305
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00016666666666666666,
+      "loss": 0.7394,
+      "step": 2310
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.000166358024691358,
+      "loss": 0.7573,
+      "step": 2315
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.00016604938271604937,
+      "loss": 0.7543,
+      "step": 2320
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.00016574074074074074,
+      "loss": 0.7709,
+      "step": 2325
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0001654320987654321,
+      "loss": 0.7372,
+      "step": 2330
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.00016512345679012342,
+      "loss": 0.7222,
+      "step": 2335
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0001648148148148148,
+      "loss": 0.7505,
+      "step": 2340
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00016450617283950616,
+      "loss": 0.7364,
+      "step": 2345
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.0001641975308641975,
+      "loss": 0.7001,
+      "step": 2350
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00016388888888888887,
+      "loss": 0.7203,
+      "step": 2355
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00016358024691358024,
+      "loss": 0.6906,
+      "step": 2360
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.0001632716049382716,
+      "loss": 0.7557,
+      "step": 2365
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00016296296296296295,
+      "loss": 0.7261,
+      "step": 2370
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.0001626543209876543,
+      "loss": 0.7182,
+      "step": 2375
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00016234567901234566,
+      "loss": 0.7557,
+      "step": 2380
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.000162037037037037,
+      "loss": 0.7251,
+      "step": 2385
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00016172839506172837,
+      "loss": 0.7551,
+      "step": 2390
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00016141975308641974,
+      "loss": 0.6976,
+      "step": 2395
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0001611111111111111,
+      "loss": 0.7336,
+      "step": 2400
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00016080246913580245,
+      "loss": 0.7023,
+      "step": 2405
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00016049382716049382,
+      "loss": 0.7315,
+      "step": 2410
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0001601851851851852,
+      "loss": 0.7092,
+      "step": 2415
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.0001598765432098765,
+      "loss": 0.7381,
+      "step": 2420
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00015956790123456788,
+      "loss": 0.7687,
+      "step": 2425
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00015925925925925924,
+      "loss": 0.7447,
+      "step": 2430
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00015895061728395059,
+      "loss": 0.7476,
+      "step": 2435
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00015864197530864196,
+      "loss": 0.7436,
+      "step": 2440
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00015833333333333332,
+      "loss": 0.7188,
+      "step": 2445
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0001580246913580247,
+      "loss": 0.7284,
+      "step": 2450
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00015771604938271603,
+      "loss": 0.739,
+      "step": 2455
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0001574074074074074,
+      "loss": 0.7108,
+      "step": 2460
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00015709876543209877,
+      "loss": 0.7101,
+      "step": 2465
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0001567901234567901,
+      "loss": 0.7433,
+      "step": 2470
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.00015648148148148146,
+      "loss": 0.7009,
+      "step": 2475
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.00015617283950617283,
+      "loss": 0.7046,
+      "step": 2480
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0001558641975308642,
+      "loss": 0.7267,
+      "step": 2485
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.00015555555555555554,
+      "loss": 0.7253,
+      "step": 2490
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.0001552469135802469,
+      "loss": 0.6932,
+      "step": 2495
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.00015493827160493827,
+      "loss": 0.7542,
+      "step": 2500
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.00015462962962962962,
+      "loss": 0.7269,
+      "step": 2505
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.00015432098765432098,
+      "loss": 0.6682,
+      "step": 2510
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.00015401234567901233,
+      "loss": 0.6625,
+      "step": 2515
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.0001537037037037037,
+      "loss": 0.6818,
+      "step": 2520
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.00015339506172839504,
+      "loss": 0.6982,
+      "step": 2525
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.0001530864197530864,
+      "loss": 0.6749,
+      "step": 2530
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.00015277777777777777,
+      "loss": 0.6552,
+      "step": 2535
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.00015246913580246912,
+      "loss": 0.702,
+      "step": 2540
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.00015216049382716049,
+      "loss": 0.6748,
+      "step": 2545
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.00015185185185185185,
+      "loss": 0.6707,
+      "step": 2550
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.00015154320987654317,
+      "loss": 0.6886,
+      "step": 2555
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.00015123456790123454,
+      "loss": 0.72,
+      "step": 2560
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.0001509259259259259,
+      "loss": 0.6959,
+      "step": 2565
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.00015061728395061728,
+      "loss": 0.6684,
+      "step": 2570
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.00015030864197530862,
+      "loss": 0.6896,
+      "step": 2575
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.00015,
+      "loss": 0.6791,
+      "step": 2580
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.00014969135802469133,
+      "loss": 0.6848,
+      "step": 2585
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.0001493827160493827,
+      "loss": 0.6471,
+      "step": 2590
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.00014907407407407407,
+      "loss": 0.6611,
+      "step": 2595
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.00014876543209876544,
+      "loss": 0.6975,
+      "step": 2600
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.00014845679012345678,
+      "loss": 0.6966,
+      "step": 2605
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.00014814814814814812,
+      "loss": 0.6894,
+      "step": 2610
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.0001478395061728395,
+      "loss": 0.681,
+      "step": 2615
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.00014753086419753086,
+      "loss": 0.6723,
+      "step": 2620
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.00014722222222222223,
+      "loss": 0.6869,
+      "step": 2625
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.00014691358024691357,
+      "loss": 0.7015,
+      "step": 2630
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.00014660493827160494,
+      "loss": 0.6932,
+      "step": 2635
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.00014629629629629628,
+      "loss": 0.6901,
+      "step": 2640
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.00014598765432098765,
+      "loss": 0.6466,
+      "step": 2645
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.000145679012345679,
+      "loss": 0.7214,
+      "step": 2650
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.00014537037037037036,
+      "loss": 0.6692,
+      "step": 2655
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.00014506172839506173,
+      "loss": 0.6935,
+      "step": 2660
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.00014475308641975307,
+      "loss": 0.6851,
+      "step": 2665
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.0001444444444444444,
+      "loss": 0.6856,
+      "step": 2670
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.00014413580246913578,
+      "loss": 0.6895,
+      "step": 2675
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.00014382716049382715,
+      "loss": 0.6562,
+      "step": 2680
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.00014351851851851852,
+      "loss": 0.6787,
+      "step": 2685
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.00014320987654320986,
+      "loss": 0.6733,
+      "step": 2690
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.00014290123456790123,
+      "loss": 0.6769,
+      "step": 2695
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.00014259259259259257,
+      "loss": 0.6543,
+      "step": 2700
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.00014228395061728394,
+      "loss": 0.6624,
+      "step": 2705
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.0001419753086419753,
+      "loss": 0.6712,
+      "step": 2710
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.00014166666666666665,
+      "loss": 0.6868,
+      "step": 2715
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.00014135802469135802,
+      "loss": 0.6584,
+      "step": 2720
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.00014104938271604936,
+      "loss": 0.6901,
+      "step": 2725
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.00014074074074074073,
+      "loss": 0.6662,
+      "step": 2730
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.0001404320987654321,
+      "loss": 0.6788,
+      "step": 2735
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.00014012345679012344,
+      "loss": 0.7102,
+      "step": 2740
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.0001398148148148148,
+      "loss": 0.6997,
+      "step": 2745
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.00013950617283950615,
+      "loss": 0.6888,
+      "step": 2750
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.00013919753086419752,
+      "loss": 0.675,
+      "step": 2755
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.0001388888888888889,
+      "loss": 0.6674,
+      "step": 2760
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.00013858024691358023,
+      "loss": 0.6982,
+      "step": 2765
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.0001382716049382716,
+      "loss": 0.6847,
+      "step": 2770
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.00013796296296296294,
+      "loss": 0.7183,
+      "step": 2775
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.0001376543209876543,
+      "loss": 0.6647,
+      "step": 2780
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.00013734567901234568,
+      "loss": 0.6794,
+      "step": 2785
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.00013703703703703702,
+      "loss": 0.6852,
+      "step": 2790
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.0001367283950617284,
+      "loss": 0.6638,
+      "step": 2795
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.00013641975308641973,
+      "loss": 0.7229,
+      "step": 2800
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.0001361111111111111,
+      "loss": 0.6804,
+      "step": 2805
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.00013580246913580244,
+      "loss": 0.7028,
+      "step": 2810
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.0001354938271604938,
+      "loss": 0.6775,
+      "step": 2815
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.00013518518518518518,
+      "loss": 0.6511,
+      "step": 2820
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.00013487654320987652,
+      "loss": 0.6952,
+      "step": 2825
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.0001345679012345679,
+      "loss": 0.6555,
+      "step": 2830
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.00013425925925925923,
+      "loss": 0.6738,
+      "step": 2835
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.0001339506172839506,
+      "loss": 0.6576,
+      "step": 2840
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.00013364197530864197,
+      "loss": 0.6679,
+      "step": 2845
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.0001333333333333333,
+      "loss": 0.6787,
+      "step": 2850
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.00013302469135802468,
+      "loss": 0.6604,
+      "step": 2855
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.00013271604938271602,
+      "loss": 0.6733,
+      "step": 2860
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.0001324074074074074,
+      "loss": 0.6829,
+      "step": 2865
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.00013209876543209876,
+      "loss": 0.6757,
+      "step": 2870
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.00013179012345679013,
+      "loss": 0.6709,
+      "step": 2875
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.00013148148148148147,
+      "loss": 0.6975,
+      "step": 2880
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.00013117283950617281,
+      "loss": 0.6719,
+      "step": 2885
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.00013086419753086418,
+      "loss": 0.664,
+      "step": 2890
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.00013055555555555555,
+      "loss": 0.6839,
+      "step": 2895
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.00013024691358024692,
+      "loss": 0.706,
+      "step": 2900
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.00012993827160493826,
+      "loss": 0.6666,
+      "step": 2905
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.0001296296296296296,
+      "loss": 0.6783,
+      "step": 2910
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.00012932098765432097,
+      "loss": 0.6746,
+      "step": 2915
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.00012901234567901234,
+      "loss": 0.6933,
+      "step": 2920
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.0001287037037037037,
+      "loss": 0.6605,
+      "step": 2925
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.00012839506172839505,
+      "loss": 0.6757,
+      "step": 2930
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.00012808641975308642,
+      "loss": 0.6385,
+      "step": 2935
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.00012777777777777776,
+      "loss": 0.6673,
+      "step": 2940
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.00012746913580246913,
+      "loss": 0.6862,
+      "step": 2945
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.00012716049382716047,
+      "loss": 0.6869,
+      "step": 2950
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.00012685185185185184,
+      "loss": 0.6392,
+      "step": 2955
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.0001265432098765432,
+      "loss": 0.6928,
+      "step": 2960
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.00012623456790123455,
+      "loss": 0.6876,
+      "step": 2965
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.0001259259259259259,
+      "loss": 0.6767,
+      "step": 2970
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.00012561728395061726,
+      "loss": 0.6828,
+      "step": 2975
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.00012530864197530863,
+      "loss": 0.6716,
+      "step": 2980
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.000125,
+      "loss": 0.6753,
+      "step": 2985
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.00012469135802469134,
+      "loss": 0.7025,
+      "step": 2990
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 0.00012438271604938269,
+      "loss": 0.6941,
+      "step": 2995
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 0.00012407407407407406,
+      "loss": 0.7042,
+      "step": 3000
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 5010,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 500,
+  "total_flos": 1.2080552599774822e+19,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

save_lora_instruct_r8_3e4_s11_1/checkpoint-3000/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6bc4247ac8c6f92ef49080d32b2520dd809daff76a4805fdafb6e454ee1c5cc3
+size 5944

save_lora_instruct_r8_3e4_s11_1/checkpoint-3000/zero_to_fp32.py

@@ -0,0 +1,587 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel()
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)

save_lora_instruct_r8_3e4_s11_1/checkpoint-3500/README.md

@@ -0,0 +1,202 @@
+---
+base_model: meta-llama/Llama-2-7b-hf
+library_name: peft
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.9.0

save_lora_instruct_r8_3e4_s11_1/checkpoint-3500/adapter_config.json

@@ -0,0 +1,35 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "meta-llama/Llama-2-7b-hf",
+  "bias": "none",
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 16,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "mora_type": 1,
+  "peft_type": "LORA",
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "k_proj",
+    "v_proj",
+    "o_proj",
+    "gate_proj",
+    "down_proj",
+    "up_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_mora": false,
+  "use_rslora": false
+}

save_lora_instruct_r8_3e4_s11_1/checkpoint-3500/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b09341566527f4be03846ece61c512bcb6e4759f0534cbfcf5f0fc9ee49d376b
+size 564342240

save_lora_instruct_r8_3e4_s11_1/checkpoint-3500/global_step3500/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b82c04a8c1b07c9320cc262e25b6b12e8956fd94b79bc54a8ba26a40f014951f
+size 239914224

save_lora_instruct_r8_3e4_s11_1/checkpoint-3500/global_step3500/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d5210780dad14c4c1676ff1c8eb22a6d9e8739c649d18383b2f617b9713bb6d
+size 40303404

save_lora_instruct_r8_3e4_s11_1/checkpoint-3500/latest

@@ -0,0 +1 @@
+global_step3500

save_lora_instruct_r8_3e4_s11_1/checkpoint-3500/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4a416fef4baf7da791f4aa05f7e025ca652472141670ffd3e3487a4df1f3a1fc
+size 14244

save_lora_instruct_r8_3e4_s11_1/checkpoint-3500/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c8366892aaa4d7d701d11c7ef8c969528988d67128ed0e7dc88a6ead516d134
+size 1064

save_lora_instruct_r8_3e4_s11_1/checkpoint-3500/trainer_state.json

@@ -0,0 +1,4221 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.3970748744815542,
+  "eval_steps": 500,
+  "global_step": 3500,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "learning_rate": 9.999999999999999e-06,
+      "loss": 1.38,
+      "step": 5
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 1.9999999999999998e-05,
+      "loss": 1.3936,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 2.9999999999999997e-05,
+      "loss": 1.3006,
+      "step": 15
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.9999999999999996e-05,
+      "loss": 1.2594,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 4.9999999999999996e-05,
+      "loss": 1.1177,
+      "step": 25
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 5.9999999999999995e-05,
+      "loss": 1.0481,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 7e-05,
+      "loss": 1.0601,
+      "step": 35
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 7.999999999999999e-05,
+      "loss": 1.0473,
+      "step": 40
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 8.999999999999999e-05,
+      "loss": 0.9523,
+      "step": 45
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 9.999999999999999e-05,
+      "loss": 0.9451,
+      "step": 50
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00010999999999999998,
+      "loss": 0.9687,
+      "step": 55
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00011999999999999999,
+      "loss": 0.9054,
+      "step": 60
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00013,
+      "loss": 0.9518,
+      "step": 65
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00014,
+      "loss": 0.8763,
+      "step": 70
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00015,
+      "loss": 0.8657,
+      "step": 75
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00015999999999999999,
+      "loss": 0.9145,
+      "step": 80
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00016999999999999999,
+      "loss": 0.8296,
+      "step": 85
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00017999999999999998,
+      "loss": 0.8709,
+      "step": 90
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00018999999999999998,
+      "loss": 0.8405,
+      "step": 95
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00019999999999999998,
+      "loss": 0.8604,
+      "step": 100
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00020999999999999998,
+      "loss": 0.8607,
+      "step": 105
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00021999999999999995,
+      "loss": 0.8376,
+      "step": 110
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00023,
+      "loss": 0.8554,
+      "step": 115
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00023999999999999998,
+      "loss": 0.8456,
+      "step": 120
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00025,
+      "loss": 0.8443,
+      "step": 125
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00026,
+      "loss": 0.8327,
+      "step": 130
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00027,
+      "loss": 0.8096,
+      "step": 135
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00028,
+      "loss": 0.7972,
+      "step": 140
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00029,
+      "loss": 0.8306,
+      "step": 145
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0003,
+      "loss": 0.8409,
+      "step": 150
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0002996913580246913,
+      "loss": 0.8094,
+      "step": 155
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00029938271604938266,
+      "loss": 0.8146,
+      "step": 160
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00029907407407407405,
+      "loss": 0.8042,
+      "step": 165
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0002987654320987654,
+      "loss": 0.7784,
+      "step": 170
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0002984567901234568,
+      "loss": 0.828,
+      "step": 175
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00029814814814814813,
+      "loss": 0.8473,
+      "step": 180
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0002978395061728395,
+      "loss": 0.8354,
+      "step": 185
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00029753086419753087,
+      "loss": 0.8246,
+      "step": 190
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00029722222222222216,
+      "loss": 0.8249,
+      "step": 195
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00029691358024691355,
+      "loss": 0.8222,
+      "step": 200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0002966049382716049,
+      "loss": 0.8369,
+      "step": 205
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00029629629629629624,
+      "loss": 0.7954,
+      "step": 210
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00029598765432098763,
+      "loss": 0.8051,
+      "step": 215
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000295679012345679,
+      "loss": 0.8081,
+      "step": 220
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00029537037037037037,
+      "loss": 0.7864,
+      "step": 225
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0002950617283950617,
+      "loss": 0.7974,
+      "step": 230
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00029475308641975306,
+      "loss": 0.8306,
+      "step": 235
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00029444444444444445,
+      "loss": 0.7792,
+      "step": 240
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00029413580246913574,
+      "loss": 0.8108,
+      "step": 245
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00029382716049382714,
+      "loss": 0.8167,
+      "step": 250
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0002935185185185185,
+      "loss": 0.8358,
+      "step": 255
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0002932098765432099,
+      "loss": 0.7948,
+      "step": 260
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0002929012345679012,
+      "loss": 0.8078,
+      "step": 265
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00029259259259259256,
+      "loss": 0.7886,
+      "step": 270
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00029228395061728395,
+      "loss": 0.7903,
+      "step": 275
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0002919753086419753,
+      "loss": 0.7892,
+      "step": 280
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00029166666666666664,
+      "loss": 0.7717,
+      "step": 285
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.000291358024691358,
+      "loss": 0.7774,
+      "step": 290
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0002910493827160494,
+      "loss": 0.8053,
+      "step": 295
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0002907407407407407,
+      "loss": 0.794,
+      "step": 300
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00029043209876543206,
+      "loss": 0.7856,
+      "step": 305
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00029012345679012345,
+      "loss": 0.7713,
+      "step": 310
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0002898148148148148,
+      "loss": 0.7805,
+      "step": 315
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00028950617283950614,
+      "loss": 0.7897,
+      "step": 320
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00028919753086419753,
+      "loss": 0.7931,
+      "step": 325
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0002888888888888888,
+      "loss": 0.7739,
+      "step": 330
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0002885802469135802,
+      "loss": 0.7749,
+      "step": 335
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00028827160493827156,
+      "loss": 0.7932,
+      "step": 340
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00028796296296296296,
+      "loss": 0.7906,
+      "step": 345
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0002876543209876543,
+      "loss": 0.8094,
+      "step": 350
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00028734567901234564,
+      "loss": 0.7932,
+      "step": 355
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00028703703703703703,
+      "loss": 0.8027,
+      "step": 360
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0002867283950617284,
+      "loss": 0.8297,
+      "step": 365
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0002864197530864197,
+      "loss": 0.8228,
+      "step": 370
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0002861111111111111,
+      "loss": 0.7997,
+      "step": 375
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00028580246913580246,
+      "loss": 0.8124,
+      "step": 380
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0002854938271604938,
+      "loss": 0.8152,
+      "step": 385
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00028518518518518514,
+      "loss": 0.7833,
+      "step": 390
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00028487654320987654,
+      "loss": 0.7884,
+      "step": 395
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0002845679012345679,
+      "loss": 0.753,
+      "step": 400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0002842592592592592,
+      "loss": 0.8055,
+      "step": 405
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0002839506172839506,
+      "loss": 0.8014,
+      "step": 410
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00028364197530864196,
+      "loss": 0.782,
+      "step": 415
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0002833333333333333,
+      "loss": 0.7733,
+      "step": 420
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0002830246913580247,
+      "loss": 0.7777,
+      "step": 425
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00028271604938271604,
+      "loss": 0.752,
+      "step": 430
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0002824074074074074,
+      "loss": 0.7963,
+      "step": 435
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0002820987654320987,
+      "loss": 0.7655,
+      "step": 440
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0002817901234567901,
+      "loss": 0.8,
+      "step": 445
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00028148148148148146,
+      "loss": 0.7465,
+      "step": 450
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0002811728395061728,
+      "loss": 0.7432,
+      "step": 455
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0002808641975308642,
+      "loss": 0.758,
+      "step": 460
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00028055555555555554,
+      "loss": 0.7757,
+      "step": 465
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0002802469135802469,
+      "loss": 0.7816,
+      "step": 470
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0002799382716049382,
+      "loss": 0.7744,
+      "step": 475
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0002796296296296296,
+      "loss": 0.7807,
+      "step": 480
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00027932098765432096,
+      "loss": 0.774,
+      "step": 485
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0002790123456790123,
+      "loss": 0.7361,
+      "step": 490
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0002787037037037037,
+      "loss": 0.7663,
+      "step": 495
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00027839506172839504,
+      "loss": 0.7532,
+      "step": 500
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0002780864197530864,
+      "loss": 0.7573,
+      "step": 505
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0002777777777777778,
+      "loss": 0.7802,
+      "step": 510
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0002774691358024691,
+      "loss": 0.7871,
+      "step": 515
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00027716049382716046,
+      "loss": 0.7734,
+      "step": 520
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0002768518518518518,
+      "loss": 0.7752,
+      "step": 525
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.0002765432098765432,
+      "loss": 0.7697,
+      "step": 530
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00027623456790123454,
+      "loss": 0.7846,
+      "step": 535
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0002759259259259259,
+      "loss": 0.7592,
+      "step": 540
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0002756172839506173,
+      "loss": 0.7734,
+      "step": 545
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0002753086419753086,
+      "loss": 0.7984,
+      "step": 550
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00027499999999999996,
+      "loss": 0.7342,
+      "step": 555
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00027469135802469136,
+      "loss": 0.7479,
+      "step": 560
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0002743827160493827,
+      "loss": 0.819,
+      "step": 565
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00027407407407407404,
+      "loss": 0.7561,
+      "step": 570
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0002737654320987654,
+      "loss": 0.7303,
+      "step": 575
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0002734567901234568,
+      "loss": 0.8001,
+      "step": 580
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0002731481481481481,
+      "loss": 0.7738,
+      "step": 585
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00027283950617283946,
+      "loss": 0.7866,
+      "step": 590
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00027253086419753086,
+      "loss": 0.7582,
+      "step": 595
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0002722222222222222,
+      "loss": 0.7645,
+      "step": 600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00027191358024691354,
+      "loss": 0.7839,
+      "step": 605
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0002716049382716049,
+      "loss": 0.7931,
+      "step": 610
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0002712962962962963,
+      "loss": 0.7692,
+      "step": 615
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0002709876543209876,
+      "loss": 0.7835,
+      "step": 620
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00027067901234567896,
+      "loss": 0.7795,
+      "step": 625
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00027037037037037036,
+      "loss": 0.7639,
+      "step": 630
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0002700617283950617,
+      "loss": 0.7705,
+      "step": 635
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00026975308641975304,
+      "loss": 0.7674,
+      "step": 640
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00026944444444444444,
+      "loss": 0.7349,
+      "step": 645
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0002691358024691358,
+      "loss": 0.8008,
+      "step": 650
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0002688271604938271,
+      "loss": 0.755,
+      "step": 655
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00026851851851851847,
+      "loss": 0.7526,
+      "step": 660
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00026820987654320986,
+      "loss": 0.7611,
+      "step": 665
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0002679012345679012,
+      "loss": 0.7621,
+      "step": 670
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00026759259259259255,
+      "loss": 0.7768,
+      "step": 675
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00026728395061728394,
+      "loss": 0.7703,
+      "step": 680
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0002669753086419753,
+      "loss": 0.7667,
+      "step": 685
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0002666666666666666,
+      "loss": 0.7712,
+      "step": 690
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.000266358024691358,
+      "loss": 0.7269,
+      "step": 695
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00026604938271604936,
+      "loss": 0.7275,
+      "step": 700
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0002657407407407407,
+      "loss": 0.7786,
+      "step": 705
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00026543209876543205,
+      "loss": 0.7808,
+      "step": 710
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00026512345679012344,
+      "loss": 0.7455,
+      "step": 715
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0002648148148148148,
+      "loss": 0.7717,
+      "step": 720
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0002645061728395061,
+      "loss": 0.737,
+      "step": 725
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0002641975308641975,
+      "loss": 0.7664,
+      "step": 730
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00026388888888888886,
+      "loss": 0.7716,
+      "step": 735
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00026358024691358026,
+      "loss": 0.7638,
+      "step": 740
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00026327160493827155,
+      "loss": 0.7777,
+      "step": 745
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00026296296296296294,
+      "loss": 0.7431,
+      "step": 750
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0002626543209876543,
+      "loss": 0.7516,
+      "step": 755
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00026234567901234563,
+      "loss": 0.831,
+      "step": 760
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.000262037037037037,
+      "loss": 0.7423,
+      "step": 765
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00026172839506172837,
+      "loss": 0.7471,
+      "step": 770
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0002614197530864197,
+      "loss": 0.7551,
+      "step": 775
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0002611111111111111,
+      "loss": 0.7208,
+      "step": 780
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00026080246913580245,
+      "loss": 0.7359,
+      "step": 785
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00026049382716049384,
+      "loss": 0.7759,
+      "step": 790
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00026018518518518513,
+      "loss": 0.7406,
+      "step": 795
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0002598765432098765,
+      "loss": 0.756,
+      "step": 800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00025956790123456787,
+      "loss": 0.7631,
+      "step": 805
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0002592592592592592,
+      "loss": 0.7885,
+      "step": 810
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0002589506172839506,
+      "loss": 0.7947,
+      "step": 815
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00025864197530864195,
+      "loss": 0.7557,
+      "step": 820
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00025833333333333334,
+      "loss": 0.7556,
+      "step": 825
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0002580246913580247,
+      "loss": 0.7472,
+      "step": 830
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.000257716049382716,
+      "loss": 0.729,
+      "step": 835
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0002574074074074074,
+      "loss": 0.7523,
+      "step": 840
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0002570987654320987,
+      "loss": 0.7505,
+      "step": 845
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0002567901234567901,
+      "loss": 0.7617,
+      "step": 850
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00025648148148148145,
+      "loss": 0.7898,
+      "step": 855
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00025617283950617284,
+      "loss": 0.7575,
+      "step": 860
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0002558641975308642,
+      "loss": 0.8316,
+      "step": 865
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00025555555555555553,
+      "loss": 0.7261,
+      "step": 870
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0002552469135802469,
+      "loss": 0.7594,
+      "step": 875
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00025493827160493826,
+      "loss": 0.7372,
+      "step": 880
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0002546296296296296,
+      "loss": 0.7498,
+      "step": 885
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00025432098765432095,
+      "loss": 0.7398,
+      "step": 890
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0002540123456790123,
+      "loss": 0.804,
+      "step": 895
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0002537037037037037,
+      "loss": 0.7241,
+      "step": 900
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00025339506172839503,
+      "loss": 0.7519,
+      "step": 905
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0002530864197530864,
+      "loss": 0.7694,
+      "step": 910
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00025277777777777777,
+      "loss": 0.7474,
+      "step": 915
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0002524691358024691,
+      "loss": 0.7278,
+      "step": 920
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0002521604938271605,
+      "loss": 0.7747,
+      "step": 925
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0002518518518518518,
+      "loss": 0.7858,
+      "step": 930
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0002515432098765432,
+      "loss": 0.7442,
+      "step": 935
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00025123456790123453,
+      "loss": 0.7446,
+      "step": 940
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0002509259259259259,
+      "loss": 0.7843,
+      "step": 945
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00025061728395061727,
+      "loss": 0.7512,
+      "step": 950
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0002503086419753086,
+      "loss": 0.7616,
+      "step": 955
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00025,
+      "loss": 0.826,
+      "step": 960
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00024969135802469135,
+      "loss": 0.7567,
+      "step": 965
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0002493827160493827,
+      "loss": 0.7754,
+      "step": 970
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0002490740740740741,
+      "loss": 0.7605,
+      "step": 975
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00024876543209876537,
+      "loss": 0.7728,
+      "step": 980
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00024845679012345677,
+      "loss": 0.7902,
+      "step": 985
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0002481481481481481,
+      "loss": 0.7622,
+      "step": 990
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0002478395061728395,
+      "loss": 0.745,
+      "step": 995
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00024753086419753085,
+      "loss": 0.7529,
+      "step": 1000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0002472222222222222,
+      "loss": 0.759,
+      "step": 1005
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0002469135802469136,
+      "loss": 0.7471,
+      "step": 1010
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00024660493827160493,
+      "loss": 0.7517,
+      "step": 1015
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00024629629629629627,
+      "loss": 0.7634,
+      "step": 1020
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0002459876543209876,
+      "loss": 0.7833,
+      "step": 1025
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.000245679012345679,
+      "loss": 0.7227,
+      "step": 1030
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00024537037037037035,
+      "loss": 0.7792,
+      "step": 1035
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002450617283950617,
+      "loss": 0.7292,
+      "step": 1040
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0002447530864197531,
+      "loss": 0.7519,
+      "step": 1045
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00024444444444444443,
+      "loss": 0.7385,
+      "step": 1050
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00024413580246913577,
+      "loss": 0.7842,
+      "step": 1055
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00024382716049382714,
+      "loss": 0.796,
+      "step": 1060
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002435185185185185,
+      "loss": 0.7479,
+      "step": 1065
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00024320987654320985,
+      "loss": 0.7531,
+      "step": 1070
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00024290123456790122,
+      "loss": 0.731,
+      "step": 1075
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0002425925925925926,
+      "loss": 0.7319,
+      "step": 1080
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00024228395061728393,
+      "loss": 0.7381,
+      "step": 1085
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024197530864197527,
+      "loss": 0.7621,
+      "step": 1090
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024166666666666664,
+      "loss": 0.771,
+      "step": 1095
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024135802469135798,
+      "loss": 0.7493,
+      "step": 1100
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024104938271604935,
+      "loss": 0.7553,
+      "step": 1105
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00024074074074074072,
+      "loss": 0.7862,
+      "step": 1110
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002404320987654321,
+      "loss": 0.7531,
+      "step": 1115
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00024012345679012343,
+      "loss": 0.7688,
+      "step": 1120
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0002398148148148148,
+      "loss": 0.7666,
+      "step": 1125
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00023950617283950617,
+      "loss": 0.7484,
+      "step": 1130
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00023919753086419748,
+      "loss": 0.8048,
+      "step": 1135
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00023888888888888885,
+      "loss": 0.757,
+      "step": 1140
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00023858024691358022,
+      "loss": 0.7838,
+      "step": 1145
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002382716049382716,
+      "loss": 0.7834,
+      "step": 1150
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00023796296296296293,
+      "loss": 0.733,
+      "step": 1155
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0002376543209876543,
+      "loss": 0.7195,
+      "step": 1160
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00023734567901234567,
+      "loss": 0.7478,
+      "step": 1165
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.000237037037037037,
+      "loss": 0.7456,
+      "step": 1170
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00023672839506172838,
+      "loss": 0.722,
+      "step": 1175
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00023641975308641975,
+      "loss": 0.7395,
+      "step": 1180
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00023611111111111112,
+      "loss": 0.7553,
+      "step": 1185
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00023580246913580243,
+      "loss": 0.7692,
+      "step": 1190
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002354938271604938,
+      "loss": 0.7396,
+      "step": 1195
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00023518518518518517,
+      "loss": 0.7214,
+      "step": 1200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0002348765432098765,
+      "loss": 0.7359,
+      "step": 1205
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00023456790123456788,
+      "loss": 0.7666,
+      "step": 1210
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00023425925925925925,
+      "loss": 0.7583,
+      "step": 1215
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002339506172839506,
+      "loss": 0.7715,
+      "step": 1220
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00023364197530864196,
+      "loss": 0.7423,
+      "step": 1225
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0002333333333333333,
+      "loss": 0.7598,
+      "step": 1230
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00023302469135802467,
+      "loss": 0.7251,
+      "step": 1235
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00023271604938271601,
+      "loss": 0.7447,
+      "step": 1240
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00023240740740740738,
+      "loss": 0.7923,
+      "step": 1245
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00023209876543209875,
+      "loss": 0.7161,
+      "step": 1250
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0002317901234567901,
+      "loss": 0.7291,
+      "step": 1255
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00023148148148148146,
+      "loss": 0.7366,
+      "step": 1260
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00023117283950617283,
+      "loss": 0.7173,
+      "step": 1265
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002308641975308642,
+      "loss": 0.7575,
+      "step": 1270
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00023055555555555552,
+      "loss": 0.7632,
+      "step": 1275
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00023024691358024688,
+      "loss": 0.7601,
+      "step": 1280
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00022993827160493825,
+      "loss": 0.7395,
+      "step": 1285
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0002296296296296296,
+      "loss": 0.7686,
+      "step": 1290
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00022932098765432096,
+      "loss": 0.7531,
+      "step": 1295
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00022901234567901233,
+      "loss": 0.7675,
+      "step": 1300
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002287037037037037,
+      "loss": 0.8103,
+      "step": 1305
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00022839506172839504,
+      "loss": 0.773,
+      "step": 1310
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0002280864197530864,
+      "loss": 0.7549,
+      "step": 1315
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00022777777777777778,
+      "loss": 0.7547,
+      "step": 1320
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0002274691358024691,
+      "loss": 0.758,
+      "step": 1325
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00022716049382716047,
+      "loss": 0.77,
+      "step": 1330
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00022685185185185183,
+      "loss": 0.7662,
+      "step": 1335
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00022654320987654318,
+      "loss": 0.744,
+      "step": 1340
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022623456790123455,
+      "loss": 0.7486,
+      "step": 1345
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022592592592592591,
+      "loss": 0.7473,
+      "step": 1350
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022561728395061728,
+      "loss": 0.7486,
+      "step": 1355
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00022530864197530862,
+      "loss": 0.7205,
+      "step": 1360
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.000225,
+      "loss": 0.7289,
+      "step": 1365
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022469135802469134,
+      "loss": 0.7586,
+      "step": 1370
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022438271604938268,
+      "loss": 0.7175,
+      "step": 1375
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022407407407407405,
+      "loss": 0.6961,
+      "step": 1380
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022376543209876542,
+      "loss": 0.7242,
+      "step": 1385
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00022345679012345678,
+      "loss": 0.7515,
+      "step": 1390
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022314814814814813,
+      "loss": 0.7522,
+      "step": 1395
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002228395061728395,
+      "loss": 0.7651,
+      "step": 1400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022253086419753086,
+      "loss": 0.6958,
+      "step": 1405
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022222222222222218,
+      "loss": 0.7336,
+      "step": 1410
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00022191358024691355,
+      "loss": 0.7636,
+      "step": 1415
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00022160493827160492,
+      "loss": 0.7583,
+      "step": 1420
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00022129629629629626,
+      "loss": 0.7605,
+      "step": 1425
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00022098765432098763,
+      "loss": 0.7109,
+      "step": 1430
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.000220679012345679,
+      "loss": 0.7224,
+      "step": 1435
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00022037037037037037,
+      "loss": 0.7724,
+      "step": 1440
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002200617283950617,
+      "loss": 0.7637,
+      "step": 1445
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021975308641975308,
+      "loss": 0.7302,
+      "step": 1450
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021944444444444444,
+      "loss": 0.7685,
+      "step": 1455
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021913580246913576,
+      "loss": 0.751,
+      "step": 1460
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00021882716049382713,
+      "loss": 0.7768,
+      "step": 1465
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002185185185185185,
+      "loss": 0.7461,
+      "step": 1470
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00021820987654320987,
+      "loss": 0.7782,
+      "step": 1475
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002179012345679012,
+      "loss": 0.7474,
+      "step": 1480
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00021759259259259258,
+      "loss": 0.7475,
+      "step": 1485
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00021728395061728395,
+      "loss": 0.7539,
+      "step": 1490
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002169753086419753,
+      "loss": 0.7476,
+      "step": 1495
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00021666666666666666,
+      "loss": 0.7382,
+      "step": 1500
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.000216358024691358,
+      "loss": 0.7305,
+      "step": 1505
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00021604938271604937,
+      "loss": 0.7455,
+      "step": 1510
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002157407407407407,
+      "loss": 0.7286,
+      "step": 1515
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00021543209876543208,
+      "loss": 0.7653,
+      "step": 1520
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00021512345679012345,
+      "loss": 0.7443,
+      "step": 1525
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0002148148148148148,
+      "loss": 0.7026,
+      "step": 1530
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00021450617283950616,
+      "loss": 0.7436,
+      "step": 1535
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00021419753086419753,
+      "loss": 0.7101,
+      "step": 1540
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00021388888888888884,
+      "loss": 0.7484,
+      "step": 1545
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002135802469135802,
+      "loss": 0.7483,
+      "step": 1550
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00021327160493827158,
+      "loss": 0.7419,
+      "step": 1555
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00021296296296296295,
+      "loss": 0.7356,
+      "step": 1560
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002126543209876543,
+      "loss": 0.7716,
+      "step": 1565
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00021234567901234566,
+      "loss": 0.734,
+      "step": 1570
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00021203703703703703,
+      "loss": 0.7268,
+      "step": 1575
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00021172839506172837,
+      "loss": 0.7133,
+      "step": 1580
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00021141975308641974,
+      "loss": 0.7575,
+      "step": 1585
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002111111111111111,
+      "loss": 0.7219,
+      "step": 1590
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00021080246913580248,
+      "loss": 0.7394,
+      "step": 1595
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0002104938271604938,
+      "loss": 0.7729,
+      "step": 1600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00021018518518518516,
+      "loss": 0.757,
+      "step": 1605
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00020987654320987653,
+      "loss": 0.7139,
+      "step": 1610
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00020956790123456787,
+      "loss": 0.7492,
+      "step": 1615
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00020925925925925924,
+      "loss": 0.7461,
+      "step": 1620
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002089506172839506,
+      "loss": 0.7276,
+      "step": 1625
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00020864197530864198,
+      "loss": 0.7725,
+      "step": 1630
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00020833333333333332,
+      "loss": 0.7623,
+      "step": 1635
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002080246913580247,
+      "loss": 0.7394,
+      "step": 1640
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00020771604938271603,
+      "loss": 0.7387,
+      "step": 1645
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00020740740740740737,
+      "loss": 0.7519,
+      "step": 1650
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00020709876543209874,
+      "loss": 0.7625,
+      "step": 1655
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002067901234567901,
+      "loss": 0.7254,
+      "step": 1660
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00020648148148148145,
+      "loss": 0.7129,
+      "step": 1665
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00020617283950617282,
+      "loss": 0.7964,
+      "step": 1670
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002058641975308642,
+      "loss": 0.7332,
+      "step": 1675
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00020555555555555556,
+      "loss": 0.7955,
+      "step": 1680
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00020524691358024687,
+      "loss": 0.7544,
+      "step": 1685
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00020493827160493824,
+      "loss": 0.7115,
+      "step": 1690
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002046296296296296,
+      "loss": 0.7434,
+      "step": 1695
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00020432098765432095,
+      "loss": 0.7382,
+      "step": 1700
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00020401234567901232,
+      "loss": 0.7452,
+      "step": 1705
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002037037037037037,
+      "loss": 0.7551,
+      "step": 1710
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00020339506172839506,
+      "loss": 0.7202,
+      "step": 1715
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002030864197530864,
+      "loss": 0.735,
+      "step": 1720
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00020277777777777777,
+      "loss": 0.7581,
+      "step": 1725
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00020246913580246914,
+      "loss": 0.7527,
+      "step": 1730
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00020216049382716045,
+      "loss": 0.7498,
+      "step": 1735
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00020185185185185182,
+      "loss": 0.735,
+      "step": 1740
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002015432098765432,
+      "loss": 0.7286,
+      "step": 1745
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00020123456790123456,
+      "loss": 0.6891,
+      "step": 1750
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002009259259259259,
+      "loss": 0.7353,
+      "step": 1755
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00020061728395061727,
+      "loss": 0.7386,
+      "step": 1760
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00020030864197530864,
+      "loss": 0.7106,
+      "step": 1765
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00019999999999999998,
+      "loss": 0.7207,
+      "step": 1770
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00019969135802469135,
+      "loss": 0.7576,
+      "step": 1775
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00019938271604938272,
+      "loss": 0.7462,
+      "step": 1780
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00019907407407407404,
+      "loss": 0.7068,
+      "step": 1785
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0001987654320987654,
+      "loss": 0.7525,
+      "step": 1790
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019845679012345677,
+      "loss": 0.7291,
+      "step": 1795
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019814814814814814,
+      "loss": 0.7226,
+      "step": 1800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019783950617283948,
+      "loss": 0.7445,
+      "step": 1805
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019753086419753085,
+      "loss": 0.7571,
+      "step": 1810
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00019722222222222222,
+      "loss": 0.7306,
+      "step": 1815
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00019691358024691354,
+      "loss": 0.7785,
+      "step": 1820
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0001966049382716049,
+      "loss": 0.7537,
+      "step": 1825
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00019629629629629627,
+      "loss": 0.7549,
+      "step": 1830
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00019598765432098764,
+      "loss": 0.732,
+      "step": 1835
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00019567901234567898,
+      "loss": 0.7159,
+      "step": 1840
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00019537037037037035,
+      "loss": 0.7549,
+      "step": 1845
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00019506172839506172,
+      "loss": 0.7581,
+      "step": 1850
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00019475308641975306,
+      "loss": 0.7559,
+      "step": 1855
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00019444444444444443,
+      "loss": 0.7626,
+      "step": 1860
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0001941358024691358,
+      "loss": 0.7423,
+      "step": 1865
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019382716049382712,
+      "loss": 0.7672,
+      "step": 1870
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019351851851851849,
+      "loss": 0.7794,
+      "step": 1875
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019320987654320985,
+      "loss": 0.7428,
+      "step": 1880
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019290123456790122,
+      "loss": 0.7462,
+      "step": 1885
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00019259259259259257,
+      "loss": 0.702,
+      "step": 1890
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00019228395061728393,
+      "loss": 0.7626,
+      "step": 1895
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0001919753086419753,
+      "loss": 0.7357,
+      "step": 1900
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00019166666666666665,
+      "loss": 0.7408,
+      "step": 1905
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00019135802469135801,
+      "loss": 0.707,
+      "step": 1910
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00019104938271604938,
+      "loss": 0.726,
+      "step": 1915
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00019074074074074073,
+      "loss": 0.7396,
+      "step": 1920
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00019043209876543207,
+      "loss": 0.7676,
+      "step": 1925
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00019012345679012344,
+      "loss": 0.7335,
+      "step": 1930
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0001898148148148148,
+      "loss": 0.7235,
+      "step": 1935
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00018950617283950615,
+      "loss": 0.7297,
+      "step": 1940
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018919753086419752,
+      "loss": 0.7079,
+      "step": 1945
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018888888888888888,
+      "loss": 0.7629,
+      "step": 1950
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018858024691358025,
+      "loss": 0.7131,
+      "step": 1955
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018827160493827157,
+      "loss": 0.7517,
+      "step": 1960
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00018796296296296294,
+      "loss": 0.7292,
+      "step": 1965
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0001876543209876543,
+      "loss": 0.7137,
+      "step": 1970
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00018734567901234565,
+      "loss": 0.725,
+      "step": 1975
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00018703703703703702,
+      "loss": 0.7444,
+      "step": 1980
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00018672839506172839,
+      "loss": 0.7366,
+      "step": 1985
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00018641975308641973,
+      "loss": 0.7329,
+      "step": 1990
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0001861111111111111,
+      "loss": 0.7037,
+      "step": 1995
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00018580246913580247,
+      "loss": 0.7186,
+      "step": 2000
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00018549382716049383,
+      "loss": 0.7322,
+      "step": 2005
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00018518518518518515,
+      "loss": 0.7639,
+      "step": 2010
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00018487654320987652,
+      "loss": 0.7426,
+      "step": 2015
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0001845679012345679,
+      "loss": 0.7648,
+      "step": 2020
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.00018425925925925923,
+      "loss": 0.714,
+      "step": 2025
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0001839506172839506,
+      "loss": 0.7091,
+      "step": 2030
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.00018364197530864197,
+      "loss": 0.7222,
+      "step": 2035
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.00018333333333333334,
+      "loss": 0.7491,
+      "step": 2040
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.00018302469135802468,
+      "loss": 0.7659,
+      "step": 2045
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.00018271604938271605,
+      "loss": 0.727,
+      "step": 2050
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.00018240740740740742,
+      "loss": 0.7416,
+      "step": 2055
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.00018209876543209873,
+      "loss": 0.697,
+      "step": 2060
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.0001817901234567901,
+      "loss": 0.7408,
+      "step": 2065
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00018148148148148147,
+      "loss": 0.7144,
+      "step": 2070
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00018117283950617284,
+      "loss": 0.7745,
+      "step": 2075
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00018086419753086418,
+      "loss": 0.7362,
+      "step": 2080
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00018055555555555555,
+      "loss": 0.7378,
+      "step": 2085
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00018024691358024692,
+      "loss": 0.7227,
+      "step": 2090
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00017993827160493826,
+      "loss": 0.7408,
+      "step": 2095
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.0001796296296296296,
+      "loss": 0.7178,
+      "step": 2100
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00017932098765432097,
+      "loss": 0.7401,
+      "step": 2105
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.0001790123456790123,
+      "loss": 0.7484,
+      "step": 2110
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00017870370370370368,
+      "loss": 0.7271,
+      "step": 2115
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00017839506172839505,
+      "loss": 0.7897,
+      "step": 2120
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00017808641975308642,
+      "loss": 0.7374,
+      "step": 2125
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00017777777777777776,
+      "loss": 0.7314,
+      "step": 2130
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00017746913580246913,
+      "loss": 0.7338,
+      "step": 2135
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.0001771604938271605,
+      "loss": 0.7536,
+      "step": 2140
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.0001768518518518518,
+      "loss": 0.7501,
+      "step": 2145
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00017654320987654318,
+      "loss": 0.7263,
+      "step": 2150
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00017623456790123455,
+      "loss": 0.7468,
+      "step": 2155
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00017592592592592592,
+      "loss": 0.7097,
+      "step": 2160
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00017561728395061726,
+      "loss": 0.7261,
+      "step": 2165
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00017530864197530863,
+      "loss": 0.7489,
+      "step": 2170
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.000175,
+      "loss": 0.7149,
+      "step": 2175
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00017469135802469134,
+      "loss": 0.7239,
+      "step": 2180
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.0001743827160493827,
+      "loss": 0.7249,
+      "step": 2185
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00017407407407407408,
+      "loss": 0.7506,
+      "step": 2190
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0001737654320987654,
+      "loss": 0.7594,
+      "step": 2195
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.00017345679012345676,
+      "loss": 0.7412,
+      "step": 2200
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.00017314814814814813,
+      "loss": 0.7219,
+      "step": 2205
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0001728395061728395,
+      "loss": 0.7655,
+      "step": 2210
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.00017253086419753084,
+      "loss": 0.735,
+      "step": 2215
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.0001722222222222222,
+      "loss": 0.7214,
+      "step": 2220
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00017191358024691358,
+      "loss": 0.7179,
+      "step": 2225
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00017160493827160492,
+      "loss": 0.7208,
+      "step": 2230
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00017129629629629626,
+      "loss": 0.7445,
+      "step": 2235
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00017098765432098763,
+      "loss": 0.7,
+      "step": 2240
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.000170679012345679,
+      "loss": 0.7571,
+      "step": 2245
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.00017037037037037034,
+      "loss": 0.7076,
+      "step": 2250
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0001700617283950617,
+      "loss": 0.7027,
+      "step": 2255
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.00016975308641975308,
+      "loss": 0.7408,
+      "step": 2260
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.00016944444444444442,
+      "loss": 0.7095,
+      "step": 2265
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0001691358024691358,
+      "loss": 0.7314,
+      "step": 2270
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.00016882716049382716,
+      "loss": 0.714,
+      "step": 2275
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.00016851851851851853,
+      "loss": 0.7442,
+      "step": 2280
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.00016820987654320984,
+      "loss": 0.7144,
+      "step": 2285
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0001679012345679012,
+      "loss": 0.7252,
+      "step": 2290
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00016759259259259258,
+      "loss": 0.7118,
+      "step": 2295
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00016728395061728392,
+      "loss": 0.7444,
+      "step": 2300
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.0001669753086419753,
+      "loss": 0.732,
+      "step": 2305
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00016666666666666666,
+      "loss": 0.7394,
+      "step": 2310
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.000166358024691358,
+      "loss": 0.7573,
+      "step": 2315
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.00016604938271604937,
+      "loss": 0.7543,
+      "step": 2320
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.00016574074074074074,
+      "loss": 0.7709,
+      "step": 2325
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0001654320987654321,
+      "loss": 0.7372,
+      "step": 2330
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.00016512345679012342,
+      "loss": 0.7222,
+      "step": 2335
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0001648148148148148,
+      "loss": 0.7505,
+      "step": 2340
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00016450617283950616,
+      "loss": 0.7364,
+      "step": 2345
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.0001641975308641975,
+      "loss": 0.7001,
+      "step": 2350
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00016388888888888887,
+      "loss": 0.7203,
+      "step": 2355
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00016358024691358024,
+      "loss": 0.6906,
+      "step": 2360
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.0001632716049382716,
+      "loss": 0.7557,
+      "step": 2365
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00016296296296296295,
+      "loss": 0.7261,
+      "step": 2370
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.0001626543209876543,
+      "loss": 0.7182,
+      "step": 2375
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00016234567901234566,
+      "loss": 0.7557,
+      "step": 2380
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.000162037037037037,
+      "loss": 0.7251,
+      "step": 2385
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00016172839506172837,
+      "loss": 0.7551,
+      "step": 2390
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00016141975308641974,
+      "loss": 0.6976,
+      "step": 2395
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0001611111111111111,
+      "loss": 0.7336,
+      "step": 2400
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00016080246913580245,
+      "loss": 0.7023,
+      "step": 2405
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00016049382716049382,
+      "loss": 0.7315,
+      "step": 2410
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0001601851851851852,
+      "loss": 0.7092,
+      "step": 2415
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.0001598765432098765,
+      "loss": 0.7381,
+      "step": 2420
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00015956790123456788,
+      "loss": 0.7687,
+      "step": 2425
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00015925925925925924,
+      "loss": 0.7447,
+      "step": 2430
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00015895061728395059,
+      "loss": 0.7476,
+      "step": 2435
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00015864197530864196,
+      "loss": 0.7436,
+      "step": 2440
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00015833333333333332,
+      "loss": 0.7188,
+      "step": 2445
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0001580246913580247,
+      "loss": 0.7284,
+      "step": 2450
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00015771604938271603,
+      "loss": 0.739,
+      "step": 2455
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0001574074074074074,
+      "loss": 0.7108,
+      "step": 2460
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00015709876543209877,
+      "loss": 0.7101,
+      "step": 2465
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0001567901234567901,
+      "loss": 0.7433,
+      "step": 2470
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.00015648148148148146,
+      "loss": 0.7009,
+      "step": 2475
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.00015617283950617283,
+      "loss": 0.7046,
+      "step": 2480
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0001558641975308642,
+      "loss": 0.7267,
+      "step": 2485
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.00015555555555555554,
+      "loss": 0.7253,
+      "step": 2490
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.0001552469135802469,
+      "loss": 0.6932,
+      "step": 2495
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.00015493827160493827,
+      "loss": 0.7542,
+      "step": 2500
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.00015462962962962962,
+      "loss": 0.7269,
+      "step": 2505
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.00015432098765432098,
+      "loss": 0.6682,
+      "step": 2510
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.00015401234567901233,
+      "loss": 0.6625,
+      "step": 2515
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.0001537037037037037,
+      "loss": 0.6818,
+      "step": 2520
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.00015339506172839504,
+      "loss": 0.6982,
+      "step": 2525
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.0001530864197530864,
+      "loss": 0.6749,
+      "step": 2530
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.00015277777777777777,
+      "loss": 0.6552,
+      "step": 2535
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.00015246913580246912,
+      "loss": 0.702,
+      "step": 2540
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.00015216049382716049,
+      "loss": 0.6748,
+      "step": 2545
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.00015185185185185185,
+      "loss": 0.6707,
+      "step": 2550
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.00015154320987654317,
+      "loss": 0.6886,
+      "step": 2555
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.00015123456790123454,
+      "loss": 0.72,
+      "step": 2560
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.0001509259259259259,
+      "loss": 0.6959,
+      "step": 2565
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.00015061728395061728,
+      "loss": 0.6684,
+      "step": 2570
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.00015030864197530862,
+      "loss": 0.6896,
+      "step": 2575
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.00015,
+      "loss": 0.6791,
+      "step": 2580
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.00014969135802469133,
+      "loss": 0.6848,
+      "step": 2585
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.0001493827160493827,
+      "loss": 0.6471,
+      "step": 2590
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.00014907407407407407,
+      "loss": 0.6611,
+      "step": 2595
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.00014876543209876544,
+      "loss": 0.6975,
+      "step": 2600
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.00014845679012345678,
+      "loss": 0.6966,
+      "step": 2605
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.00014814814814814812,
+      "loss": 0.6894,
+      "step": 2610
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.0001478395061728395,
+      "loss": 0.681,
+      "step": 2615
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.00014753086419753086,
+      "loss": 0.6723,
+      "step": 2620
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.00014722222222222223,
+      "loss": 0.6869,
+      "step": 2625
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.00014691358024691357,
+      "loss": 0.7015,
+      "step": 2630
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.00014660493827160494,
+      "loss": 0.6932,
+      "step": 2635
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.00014629629629629628,
+      "loss": 0.6901,
+      "step": 2640
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.00014598765432098765,
+      "loss": 0.6466,
+      "step": 2645
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.000145679012345679,
+      "loss": 0.7214,
+      "step": 2650
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.00014537037037037036,
+      "loss": 0.6692,
+      "step": 2655
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.00014506172839506173,
+      "loss": 0.6935,
+      "step": 2660
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.00014475308641975307,
+      "loss": 0.6851,
+      "step": 2665
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.0001444444444444444,
+      "loss": 0.6856,
+      "step": 2670
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.00014413580246913578,
+      "loss": 0.6895,
+      "step": 2675
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.00014382716049382715,
+      "loss": 0.6562,
+      "step": 2680
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.00014351851851851852,
+      "loss": 0.6787,
+      "step": 2685
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.00014320987654320986,
+      "loss": 0.6733,
+      "step": 2690
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.00014290123456790123,
+      "loss": 0.6769,
+      "step": 2695
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.00014259259259259257,
+      "loss": 0.6543,
+      "step": 2700
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.00014228395061728394,
+      "loss": 0.6624,
+      "step": 2705
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.0001419753086419753,
+      "loss": 0.6712,
+      "step": 2710
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.00014166666666666665,
+      "loss": 0.6868,
+      "step": 2715
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.00014135802469135802,
+      "loss": 0.6584,
+      "step": 2720
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.00014104938271604936,
+      "loss": 0.6901,
+      "step": 2725
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.00014074074074074073,
+      "loss": 0.6662,
+      "step": 2730
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.0001404320987654321,
+      "loss": 0.6788,
+      "step": 2735
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.00014012345679012344,
+      "loss": 0.7102,
+      "step": 2740
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.0001398148148148148,
+      "loss": 0.6997,
+      "step": 2745
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.00013950617283950615,
+      "loss": 0.6888,
+      "step": 2750
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.00013919753086419752,
+      "loss": 0.675,
+      "step": 2755
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.0001388888888888889,
+      "loss": 0.6674,
+      "step": 2760
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.00013858024691358023,
+      "loss": 0.6982,
+      "step": 2765
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.0001382716049382716,
+      "loss": 0.6847,
+      "step": 2770
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.00013796296296296294,
+      "loss": 0.7183,
+      "step": 2775
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.0001376543209876543,
+      "loss": 0.6647,
+      "step": 2780
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.00013734567901234568,
+      "loss": 0.6794,
+      "step": 2785
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.00013703703703703702,
+      "loss": 0.6852,
+      "step": 2790
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.0001367283950617284,
+      "loss": 0.6638,
+      "step": 2795
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.00013641975308641973,
+      "loss": 0.7229,
+      "step": 2800
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.0001361111111111111,
+      "loss": 0.6804,
+      "step": 2805
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.00013580246913580244,
+      "loss": 0.7028,
+      "step": 2810
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.0001354938271604938,
+      "loss": 0.6775,
+      "step": 2815
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.00013518518518518518,
+      "loss": 0.6511,
+      "step": 2820
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.00013487654320987652,
+      "loss": 0.6952,
+      "step": 2825
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.0001345679012345679,
+      "loss": 0.6555,
+      "step": 2830
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.00013425925925925923,
+      "loss": 0.6738,
+      "step": 2835
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.0001339506172839506,
+      "loss": 0.6576,
+      "step": 2840
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.00013364197530864197,
+      "loss": 0.6679,
+      "step": 2845
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.0001333333333333333,
+      "loss": 0.6787,
+      "step": 2850
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.00013302469135802468,
+      "loss": 0.6604,
+      "step": 2855
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.00013271604938271602,
+      "loss": 0.6733,
+      "step": 2860
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.0001324074074074074,
+      "loss": 0.6829,
+      "step": 2865
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.00013209876543209876,
+      "loss": 0.6757,
+      "step": 2870
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.00013179012345679013,
+      "loss": 0.6709,
+      "step": 2875
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.00013148148148148147,
+      "loss": 0.6975,
+      "step": 2880
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.00013117283950617281,
+      "loss": 0.6719,
+      "step": 2885
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.00013086419753086418,
+      "loss": 0.664,
+      "step": 2890
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.00013055555555555555,
+      "loss": 0.6839,
+      "step": 2895
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.00013024691358024692,
+      "loss": 0.706,
+      "step": 2900
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.00012993827160493826,
+      "loss": 0.6666,
+      "step": 2905
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.0001296296296296296,
+      "loss": 0.6783,
+      "step": 2910
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.00012932098765432097,
+      "loss": 0.6746,
+      "step": 2915
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.00012901234567901234,
+      "loss": 0.6933,
+      "step": 2920
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.0001287037037037037,
+      "loss": 0.6605,
+      "step": 2925
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.00012839506172839505,
+      "loss": 0.6757,
+      "step": 2930
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.00012808641975308642,
+      "loss": 0.6385,
+      "step": 2935
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.00012777777777777776,
+      "loss": 0.6673,
+      "step": 2940
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.00012746913580246913,
+      "loss": 0.6862,
+      "step": 2945
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.00012716049382716047,
+      "loss": 0.6869,
+      "step": 2950
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.00012685185185185184,
+      "loss": 0.6392,
+      "step": 2955
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.0001265432098765432,
+      "loss": 0.6928,
+      "step": 2960
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.00012623456790123455,
+      "loss": 0.6876,
+      "step": 2965
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.0001259259259259259,
+      "loss": 0.6767,
+      "step": 2970
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.00012561728395061726,
+      "loss": 0.6828,
+      "step": 2975
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.00012530864197530863,
+      "loss": 0.6716,
+      "step": 2980
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.000125,
+      "loss": 0.6753,
+      "step": 2985
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.00012469135802469134,
+      "loss": 0.7025,
+      "step": 2990
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 0.00012438271604938269,
+      "loss": 0.6941,
+      "step": 2995
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 0.00012407407407407406,
+      "loss": 0.7042,
+      "step": 3000
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 0.00012376543209876542,
+      "loss": 0.6651,
+      "step": 3005
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 0.0001234567901234568,
+      "loss": 0.6793,
+      "step": 3010
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 0.00012314814814814813,
+      "loss": 0.6473,
+      "step": 3015
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 0.0001228395061728395,
+      "loss": 0.6683,
+      "step": 3020
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 0.00012253086419753085,
+      "loss": 0.6697,
+      "step": 3025
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 0.00012222222222222221,
+      "loss": 0.6639,
+      "step": 3030
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 0.00012191358024691357,
+      "loss": 0.6642,
+      "step": 3035
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 0.00012160493827160493,
+      "loss": 0.6654,
+      "step": 3040
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 0.0001212962962962963,
+      "loss": 0.712,
+      "step": 3045
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 0.00012098765432098764,
+      "loss": 0.6702,
+      "step": 3050
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 0.00012067901234567899,
+      "loss": 0.6857,
+      "step": 3055
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 0.00012037037037037036,
+      "loss": 0.6755,
+      "step": 3060
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 0.00012006172839506172,
+      "loss": 0.6617,
+      "step": 3065
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 0.00011975308641975308,
+      "loss": 0.6794,
+      "step": 3070
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 0.00011944444444444443,
+      "loss": 0.6807,
+      "step": 3075
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 0.0001191358024691358,
+      "loss": 0.661,
+      "step": 3080
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 0.00011882716049382715,
+      "loss": 0.691,
+      "step": 3085
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 0.0001185185185185185,
+      "loss": 0.6799,
+      "step": 3090
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 0.00011820987654320988,
+      "loss": 0.6598,
+      "step": 3095
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 0.00011790123456790122,
+      "loss": 0.7039,
+      "step": 3100
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 0.00011759259259259259,
+      "loss": 0.6943,
+      "step": 3105
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 0.00011728395061728394,
+      "loss": 0.7052,
+      "step": 3110
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 0.0001169753086419753,
+      "loss": 0.6884,
+      "step": 3115
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 0.00011666666666666665,
+      "loss": 0.711,
+      "step": 3120
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 0.00011635802469135801,
+      "loss": 0.702,
+      "step": 3125
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 0.00011604938271604938,
+      "loss": 0.6905,
+      "step": 3130
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 0.00011574074074074073,
+      "loss": 0.6433,
+      "step": 3135
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 0.0001154320987654321,
+      "loss": 0.6643,
+      "step": 3140
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 0.00011512345679012344,
+      "loss": 0.6841,
+      "step": 3145
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 0.0001148148148148148,
+      "loss": 0.6423,
+      "step": 3150
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 0.00011450617283950617,
+      "loss": 0.651,
+      "step": 3155
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 0.00011419753086419752,
+      "loss": 0.6819,
+      "step": 3160
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 0.00011388888888888889,
+      "loss": 0.7022,
+      "step": 3165
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 0.00011358024691358023,
+      "loss": 0.6715,
+      "step": 3170
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 0.00011327160493827159,
+      "loss": 0.6813,
+      "step": 3175
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 0.00011296296296296296,
+      "loss": 0.6756,
+      "step": 3180
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 0.00011265432098765431,
+      "loss": 0.7021,
+      "step": 3185
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 0.00011234567901234567,
+      "loss": 0.698,
+      "step": 3190
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 0.00011203703703703702,
+      "loss": 0.6465,
+      "step": 3195
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 0.00011172839506172839,
+      "loss": 0.6546,
+      "step": 3200
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 0.00011141975308641975,
+      "loss": 0.6758,
+      "step": 3205
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 0.00011111111111111109,
+      "loss": 0.6447,
+      "step": 3210
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 0.00011080246913580246,
+      "loss": 0.6741,
+      "step": 3215
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 0.00011049382716049381,
+      "loss": 0.6846,
+      "step": 3220
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 0.00011018518518518518,
+      "loss": 0.6573,
+      "step": 3225
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 0.00010987654320987654,
+      "loss": 0.6738,
+      "step": 3230
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 0.00010956790123456788,
+      "loss": 0.6782,
+      "step": 3235
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 0.00010925925925925925,
+      "loss": 0.6815,
+      "step": 3240
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 0.0001089506172839506,
+      "loss": 0.6714,
+      "step": 3245
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 0.00010864197530864197,
+      "loss": 0.6839,
+      "step": 3250
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 0.00010833333333333333,
+      "loss": 0.6564,
+      "step": 3255
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 0.00010802469135802468,
+      "loss": 0.6524,
+      "step": 3260
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 0.00010771604938271604,
+      "loss": 0.6636,
+      "step": 3265
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 0.0001074074074074074,
+      "loss": 0.66,
+      "step": 3270
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 0.00010709876543209876,
+      "loss": 0.7089,
+      "step": 3275
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 0.0001067901234567901,
+      "loss": 0.6692,
+      "step": 3280
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 0.00010648148148148147,
+      "loss": 0.6871,
+      "step": 3285
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 0.00010617283950617283,
+      "loss": 0.6749,
+      "step": 3290
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 0.00010586419753086418,
+      "loss": 0.684,
+      "step": 3295
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 0.00010555555555555555,
+      "loss": 0.6803,
+      "step": 3300
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 0.0001052469135802469,
+      "loss": 0.6932,
+      "step": 3305
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 0.00010493827160493826,
+      "loss": 0.6542,
+      "step": 3310
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 0.00010462962962962962,
+      "loss": 0.6494,
+      "step": 3315
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 0.00010432098765432099,
+      "loss": 0.6804,
+      "step": 3320
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 0.00010401234567901234,
+      "loss": 0.6709,
+      "step": 3325
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 0.00010370370370370369,
+      "loss": 0.7199,
+      "step": 3330
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 0.00010339506172839506,
+      "loss": 0.6677,
+      "step": 3335
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 0.00010308641975308641,
+      "loss": 0.6657,
+      "step": 3340
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 0.00010277777777777778,
+      "loss": 0.6811,
+      "step": 3345
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 0.00010246913580246912,
+      "loss": 0.6635,
+      "step": 3350
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 0.00010216049382716048,
+      "loss": 0.6696,
+      "step": 3355
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 0.00010185185185185185,
+      "loss": 0.6926,
+      "step": 3360
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 0.0001015432098765432,
+      "loss": 0.6638,
+      "step": 3365
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 0.00010123456790123457,
+      "loss": 0.6518,
+      "step": 3370
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 0.00010092592592592591,
+      "loss": 0.6521,
+      "step": 3375
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 0.00010061728395061728,
+      "loss": 0.6809,
+      "step": 3380
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 0.00010030864197530864,
+      "loss": 0.6887,
+      "step": 3385
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 9.999999999999999e-05,
+      "loss": 0.6924,
+      "step": 3390
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 9.969135802469136e-05,
+      "loss": 0.6674,
+      "step": 3395
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 9.93827160493827e-05,
+      "loss": 0.6908,
+      "step": 3400
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 9.907407407407407e-05,
+      "loss": 0.6959,
+      "step": 3405
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 9.876543209876543e-05,
+      "loss": 0.7068,
+      "step": 3410
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 9.845679012345677e-05,
+      "loss": 0.6694,
+      "step": 3415
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 9.814814814814814e-05,
+      "loss": 0.6214,
+      "step": 3420
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 9.783950617283949e-05,
+      "loss": 0.6534,
+      "step": 3425
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 9.753086419753086e-05,
+      "loss": 0.6708,
+      "step": 3430
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 9.722222222222222e-05,
+      "loss": 0.6624,
+      "step": 3435
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 9.691358024691356e-05,
+      "loss": 0.676,
+      "step": 3440
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 9.660493827160493e-05,
+      "loss": 0.6516,
+      "step": 3445
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 9.629629629629628e-05,
+      "loss": 0.6688,
+      "step": 3450
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 9.598765432098765e-05,
+      "loss": 0.6658,
+      "step": 3455
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 9.567901234567901e-05,
+      "loss": 0.6582,
+      "step": 3460
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 9.537037037037036e-05,
+      "loss": 0.6609,
+      "step": 3465
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 9.506172839506172e-05,
+      "loss": 0.6827,
+      "step": 3470
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 9.475308641975307e-05,
+      "loss": 0.6683,
+      "step": 3475
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 9.444444444444444e-05,
+      "loss": 0.6609,
+      "step": 3480
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 9.413580246913578e-05,
+      "loss": 0.6608,
+      "step": 3485
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 9.382716049382715e-05,
+      "loss": 0.6539,
+      "step": 3490
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 9.351851851851851e-05,
+      "loss": 0.6515,
+      "step": 3495
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 9.320987654320986e-05,
+      "loss": 0.6662,
+      "step": 3500
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 5010,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 500,
+  "total_flos": 1.408889273989464e+19,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

save_lora_instruct_r8_3e4_s11_1/checkpoint-3500/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6bc4247ac8c6f92ef49080d32b2520dd809daff76a4805fdafb6e454ee1c5cc3
+size 5944

save_lora_instruct_r8_3e4_s11_1/checkpoint-3500/zero_to_fp32.py

@@ -0,0 +1,587 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel()
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)

save_lora_instruct_r8_3e4_s11_1/checkpoint-4000/README.md

@@ -0,0 +1,202 @@
+---
+base_model: meta-llama/Llama-2-7b-hf
+library_name: peft
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.9.0

save_lora_instruct_r8_3e4_s11_1/checkpoint-4000/adapter_config.json

@@ -0,0 +1,35 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "meta-llama/Llama-2-7b-hf",
+  "bias": "none",
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 16,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "mora_type": 1,
+  "peft_type": "LORA",
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "k_proj",
+    "v_proj",
+    "o_proj",
+    "gate_proj",
+    "down_proj",
+    "up_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_mora": false,
+  "use_rslora": false
+}

save_lora_instruct_r8_3e4_s11_1/checkpoint-4000/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:41f677c1f3fd50c1932f41bc8ffdbfd5730d902f8dd993784b4c041e5f0e0ff0
+size 564342240