Upload 91 files

save_lora_instruct_r8_4e4_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_4e4_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_4e4_s11_1/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b877725910cf2bc993f85d0284ffc64a6a01cabf19935b768e3986df210dc2c2
+size 564342240

save_lora_instruct_r8_4e4_s11_1/checkpoint-1500/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_4e4_s11_1/checkpoint-1500/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_4e4_s11_1/checkpoint-1500/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:148b651127bad36ecafe1696dd4410a4dbc7f78c3d1729012f246f0217678672
+size 564342240

save_lora_instruct_r8_4e4_s11_1/checkpoint-1500/global_step1500/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:dc73040c87dbff03438699f1812dbe80616e5733495eebe76f36c37042b2f749
+size 239914224

save_lora_instruct_r8_4e4_s11_1/checkpoint-1500/global_step1500/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7e9748c0d074063aa0be54e4a8a370e66d0671ca8867acc5ce413e208beb76bd
+size 40303404

save_lora_instruct_r8_4e4_s11_1/checkpoint-1500/latest

@@ -0,0 +1 @@
+global_step1500

save_lora_instruct_r8_4e4_s11_1/checkpoint-1500/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb1ba1ceb33f795be7ee4e07148f4508e4f6b4ae91adf50fdaf1077b1ae47da8
+size 14244

save_lora_instruct_r8_4e4_s11_1/checkpoint-1500/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5f4641faf2b17520c52e00d1eae4553db1b75e4c42cd080785063e0d358be92
+size 1064

save_lora_instruct_r8_4e4_s11_1/checkpoint-1500/trainer_state.json

@@ -0,0 +1,1821 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.598746374777809,
+  "eval_steps": 500,
+  "global_step": 1500,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "learning_rate": 1.3333333333333333e-05,
+      "loss": 1.3798,
+      "step": 5
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 2.6666666666666667e-05,
+      "loss": 1.3917,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 4e-05,
+      "loss": 1.2862,
+      "step": 15
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 5.333333333333333e-05,
+      "loss": 1.2046,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 6.666666666666667e-05,
+      "loss": 1.0634,
+      "step": 25
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 8e-05,
+      "loss": 1.0273,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.333333333333334e-05,
+      "loss": 1.0472,
+      "step": 35
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00010666666666666667,
+      "loss": 1.0301,
+      "step": 40
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00012,
+      "loss": 0.9335,
+      "step": 45
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00013333333333333334,
+      "loss": 0.921,
+      "step": 50
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00014666666666666666,
+      "loss": 0.9506,
+      "step": 55
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00016,
+      "loss": 0.8876,
+      "step": 60
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00017333333333333334,
+      "loss": 0.9335,
+      "step": 65
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0001866666666666667,
+      "loss": 0.8554,
+      "step": 70
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0002,
+      "loss": 0.8537,
+      "step": 75
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00021333333333333333,
+      "loss": 0.9033,
+      "step": 80
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00022666666666666668,
+      "loss": 0.8203,
+      "step": 85
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00024,
+      "loss": 0.8611,
+      "step": 90
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00025333333333333333,
+      "loss": 0.8303,
+      "step": 95
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0002666666666666667,
+      "loss": 0.8528,
+      "step": 100
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00028,
+      "loss": 0.8537,
+      "step": 105
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0002933333333333333,
+      "loss": 0.8286,
+      "step": 110
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0003066666666666667,
+      "loss": 0.8449,
+      "step": 115
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00032,
+      "loss": 0.8377,
+      "step": 120
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0003333333333333334,
+      "loss": 0.8419,
+      "step": 125
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00034666666666666667,
+      "loss": 0.8265,
+      "step": 130
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00036,
+      "loss": 0.8037,
+      "step": 135
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0003733333333333334,
+      "loss": 0.7924,
+      "step": 140
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00038666666666666667,
+      "loss": 0.8245,
+      "step": 145
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004,
+      "loss": 0.8388,
+      "step": 150
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0003995884773662552,
+      "loss": 0.8066,
+      "step": 155
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0003991769547325103,
+      "loss": 0.8107,
+      "step": 160
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00039876543209876544,
+      "loss": 0.8005,
+      "step": 165
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0003983539094650206,
+      "loss": 0.7727,
+      "step": 170
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00039794238683127576,
+      "loss": 0.8276,
+      "step": 175
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00039753086419753086,
+      "loss": 0.8464,
+      "step": 180
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000397119341563786,
+      "loss": 0.8348,
+      "step": 185
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0003967078189300412,
+      "loss": 0.8228,
+      "step": 190
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0003962962962962963,
+      "loss": 0.8254,
+      "step": 195
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00039588477366255144,
+      "loss": 0.817,
+      "step": 200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0003954732510288066,
+      "loss": 0.8365,
+      "step": 205
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0003950617283950617,
+      "loss": 0.793,
+      "step": 210
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00039465020576131686,
+      "loss": 0.8014,
+      "step": 215
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000394238683127572,
+      "loss": 0.8049,
+      "step": 220
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0003938271604938272,
+      "loss": 0.7841,
+      "step": 225
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0003934156378600823,
+      "loss": 0.7977,
+      "step": 230
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00039300411522633744,
+      "loss": 0.8299,
+      "step": 235
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0003925925925925926,
+      "loss": 0.7729,
+      "step": 240
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00039218106995884776,
+      "loss": 0.8082,
+      "step": 245
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0003917695473251029,
+      "loss": 0.8173,
+      "step": 250
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000391358024691358,
+      "loss": 0.8379,
+      "step": 255
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0003909465020576132,
+      "loss": 0.7962,
+      "step": 260
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00039053497942386834,
+      "loss": 0.8067,
+      "step": 265
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0003901234567901235,
+      "loss": 0.7915,
+      "step": 270
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00038971193415637866,
+      "loss": 0.7905,
+      "step": 275
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00038930041152263376,
+      "loss": 0.7862,
+      "step": 280
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0003888888888888889,
+      "loss": 0.7709,
+      "step": 285
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0003884773662551441,
+      "loss": 0.7773,
+      "step": 290
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00038806584362139924,
+      "loss": 0.8068,
+      "step": 295
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00038765432098765434,
+      "loss": 0.7962,
+      "step": 300
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0003872427983539095,
+      "loss": 0.7894,
+      "step": 305
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00038683127572016466,
+      "loss": 0.7705,
+      "step": 310
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00038641975308641976,
+      "loss": 0.7882,
+      "step": 315
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0003860082304526749,
+      "loss": 0.7865,
+      "step": 320
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0003855967078189301,
+      "loss": 0.7939,
+      "step": 325
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0003851851851851852,
+      "loss": 0.7724,
+      "step": 330
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00038477366255144034,
+      "loss": 0.7771,
+      "step": 335
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0003843621399176955,
+      "loss": 0.7941,
+      "step": 340
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00038395061728395066,
+      "loss": 0.7822,
+      "step": 345
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00038353909465020577,
+      "loss": 0.8143,
+      "step": 350
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0003831275720164609,
+      "loss": 0.795,
+      "step": 355
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0003827160493827161,
+      "loss": 0.8049,
+      "step": 360
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0003823045267489712,
+      "loss": 0.8309,
+      "step": 365
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00038189300411522635,
+      "loss": 0.8314,
+      "step": 370
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0003814814814814815,
+      "loss": 0.801,
+      "step": 375
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00038106995884773666,
+      "loss": 0.8084,
+      "step": 380
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00038065843621399177,
+      "loss": 0.8129,
+      "step": 385
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0003802469135802469,
+      "loss": 0.7859,
+      "step": 390
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0003798353909465021,
+      "loss": 0.7874,
+      "step": 395
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0003794238683127572,
+      "loss": 0.7548,
+      "step": 400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00037901234567901235,
+      "loss": 0.8092,
+      "step": 405
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0003786008230452675,
+      "loss": 0.8041,
+      "step": 410
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00037818930041152266,
+      "loss": 0.777,
+      "step": 415
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00037777777777777777,
+      "loss": 0.7699,
+      "step": 420
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00037736625514403293,
+      "loss": 0.781,
+      "step": 425
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0003769547325102881,
+      "loss": 0.7504,
+      "step": 430
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0003765432098765432,
+      "loss": 0.7974,
+      "step": 435
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00037613168724279835,
+      "loss": 0.7663,
+      "step": 440
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0003757201646090535,
+      "loss": 0.8013,
+      "step": 445
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00037530864197530867,
+      "loss": 0.7453,
+      "step": 450
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00037489711934156377,
+      "loss": 0.7425,
+      "step": 455
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00037448559670781893,
+      "loss": 0.7582,
+      "step": 460
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0003740740740740741,
+      "loss": 0.7801,
+      "step": 465
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00037366255144032925,
+      "loss": 0.7809,
+      "step": 470
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0003732510288065844,
+      "loss": 0.7749,
+      "step": 475
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00037283950617283956,
+      "loss": 0.7807,
+      "step": 480
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00037242798353909467,
+      "loss": 0.7715,
+      "step": 485
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0003720164609053498,
+      "loss": 0.7356,
+      "step": 490
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.000371604938271605,
+      "loss": 0.7674,
+      "step": 495
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00037119341563786014,
+      "loss": 0.7542,
+      "step": 500
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00037078189300411525,
+      "loss": 0.7583,
+      "step": 505
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0003703703703703704,
+      "loss": 0.7767,
+      "step": 510
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00036995884773662556,
+      "loss": 0.7853,
+      "step": 515
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00036954732510288067,
+      "loss": 0.7733,
+      "step": 520
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00036913580246913583,
+      "loss": 0.7758,
+      "step": 525
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.000368724279835391,
+      "loss": 0.7698,
+      "step": 530
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00036831275720164614,
+      "loss": 0.7805,
+      "step": 535
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00036790123456790125,
+      "loss": 0.7573,
+      "step": 540
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003674897119341564,
+      "loss": 0.7709,
+      "step": 545
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00036707818930041157,
+      "loss": 0.7994,
+      "step": 550
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00036666666666666667,
+      "loss": 0.7363,
+      "step": 555
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00036625514403292183,
+      "loss": 0.7478,
+      "step": 560
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.000365843621399177,
+      "loss": 0.8188,
+      "step": 565
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003654320987654321,
+      "loss": 0.7524,
+      "step": 570
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00036502057613168725,
+      "loss": 0.7319,
+      "step": 575
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003646090534979424,
+      "loss": 0.7915,
+      "step": 580
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00036419753086419757,
+      "loss": 0.7722,
+      "step": 585
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00036378600823045267,
+      "loss": 0.7842,
+      "step": 590
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00036337448559670783,
+      "loss": 0.7591,
+      "step": 595
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.000362962962962963,
+      "loss": 0.7653,
+      "step": 600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003625514403292181,
+      "loss": 0.7838,
+      "step": 605
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00036213991769547325,
+      "loss": 0.7937,
+      "step": 610
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003617283950617284,
+      "loss": 0.77,
+      "step": 615
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00036131687242798357,
+      "loss": 0.7899,
+      "step": 620
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003609053497942387,
+      "loss": 0.7815,
+      "step": 625
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00036049382716049383,
+      "loss": 0.7648,
+      "step": 630
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.000360082304526749,
+      "loss": 0.7716,
+      "step": 635
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003596707818930041,
+      "loss": 0.7708,
+      "step": 640
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00035925925925925925,
+      "loss": 0.7306,
+      "step": 645
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003588477366255144,
+      "loss": 0.8007,
+      "step": 650
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003584362139917695,
+      "loss": 0.7551,
+      "step": 655
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003580246913580247,
+      "loss": 0.754,
+      "step": 660
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00035761316872427983,
+      "loss": 0.7599,
+      "step": 665
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.000357201646090535,
+      "loss": 0.7637,
+      "step": 670
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00035679012345679015,
+      "loss": 0.7766,
+      "step": 675
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003563786008230453,
+      "loss": 0.7727,
+      "step": 680
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003559670781893004,
+      "loss": 0.7675,
+      "step": 685
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00035555555555555557,
+      "loss": 0.7707,
+      "step": 690
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00035514403292181073,
+      "loss": 0.7221,
+      "step": 695
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003547325102880659,
+      "loss": 0.7335,
+      "step": 700
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00035432098765432105,
+      "loss": 0.781,
+      "step": 705
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00035390946502057615,
+      "loss": 0.7845,
+      "step": 710
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003534979423868313,
+      "loss": 0.7467,
+      "step": 715
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035308641975308647,
+      "loss": 0.7735,
+      "step": 720
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003526748971193416,
+      "loss": 0.7368,
+      "step": 725
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035226337448559673,
+      "loss": 0.7678,
+      "step": 730
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003518518518518519,
+      "loss": 0.7756,
+      "step": 735
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035144032921810705,
+      "loss": 0.7635,
+      "step": 740
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035102880658436215,
+      "loss": 0.7757,
+      "step": 745
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003506172839506173,
+      "loss": 0.7442,
+      "step": 750
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035020576131687247,
+      "loss": 0.7584,
+      "step": 755
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003497942386831276,
+      "loss": 0.8333,
+      "step": 760
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034938271604938273,
+      "loss": 0.7459,
+      "step": 765
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003489711934156379,
+      "loss": 0.7499,
+      "step": 770
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.000348559670781893,
+      "loss": 0.7571,
+      "step": 775
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034814814814814816,
+      "loss": 0.7243,
+      "step": 780
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003477366255144033,
+      "loss": 0.7443,
+      "step": 785
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003473251028806585,
+      "loss": 0.7759,
+      "step": 790
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003469135802469136,
+      "loss": 0.7452,
+      "step": 795
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034650205761316874,
+      "loss": 0.7578,
+      "step": 800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003460905349794239,
+      "loss": 0.7634,
+      "step": 805
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.000345679012345679,
+      "loss": 0.7925,
+      "step": 810
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00034526748971193416,
+      "loss": 0.7966,
+      "step": 815
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003448559670781893,
+      "loss": 0.7601,
+      "step": 820
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003444444444444445,
+      "loss": 0.7617,
+      "step": 825
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003440329218106996,
+      "loss": 0.7526,
+      "step": 830
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00034362139917695474,
+      "loss": 0.7297,
+      "step": 835
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003432098765432099,
+      "loss": 0.7549,
+      "step": 840
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.000342798353909465,
+      "loss": 0.7517,
+      "step": 845
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00034238683127572016,
+      "loss": 0.7626,
+      "step": 850
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003419753086419753,
+      "loss": 0.7987,
+      "step": 855
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003415637860082305,
+      "loss": 0.7573,
+      "step": 860
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003411522633744856,
+      "loss": 0.8367,
+      "step": 865
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00034074074074074074,
+      "loss": 0.7258,
+      "step": 870
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003403292181069959,
+      "loss": 0.7632,
+      "step": 875
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00033991769547325106,
+      "loss": 0.7379,
+      "step": 880
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00033950617283950616,
+      "loss": 0.7529,
+      "step": 885
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003390946502057613,
+      "loss": 0.7391,
+      "step": 890
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003386831275720165,
+      "loss": 0.8061,
+      "step": 895
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00033827160493827164,
+      "loss": 0.7288,
+      "step": 900
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003378600823045268,
+      "loss": 0.7493,
+      "step": 905
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00033744855967078195,
+      "loss": 0.7702,
+      "step": 910
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00033703703703703706,
+      "loss": 0.7538,
+      "step": 915
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003366255144032922,
+      "loss": 0.728,
+      "step": 920
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003362139917695474,
+      "loss": 0.7776,
+      "step": 925
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003358024691358025,
+      "loss": 0.7883,
+      "step": 930
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00033539094650205764,
+      "loss": 0.7471,
+      "step": 935
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003349794238683128,
+      "loss": 0.7464,
+      "step": 940
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00033456790123456795,
+      "loss": 0.7885,
+      "step": 945
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00033415637860082306,
+      "loss": 0.7509,
+      "step": 950
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003337448559670782,
+      "loss": 0.7681,
+      "step": 955
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003333333333333334,
+      "loss": 0.8246,
+      "step": 960
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003329218106995885,
+      "loss": 0.7604,
+      "step": 965
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00033251028806584364,
+      "loss": 0.7755,
+      "step": 970
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003320987654320988,
+      "loss": 0.7641,
+      "step": 975
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003316872427983539,
+      "loss": 0.7708,
+      "step": 980
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00033127572016460906,
+      "loss": 0.7936,
+      "step": 985
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003308641975308642,
+      "loss": 0.7641,
+      "step": 990
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003304526748971194,
+      "loss": 0.7479,
+      "step": 995
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003300411522633745,
+      "loss": 0.7547,
+      "step": 1000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00032962962962962964,
+      "loss": 0.762,
+      "step": 1005
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003292181069958848,
+      "loss": 0.7472,
+      "step": 1010
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003288065843621399,
+      "loss": 0.7542,
+      "step": 1015
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00032839506172839506,
+      "loss": 0.7669,
+      "step": 1020
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003279835390946502,
+      "loss": 0.7812,
+      "step": 1025
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003275720164609054,
+      "loss": 0.7273,
+      "step": 1030
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003271604938271605,
+      "loss": 0.7822,
+      "step": 1035
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00032674897119341564,
+      "loss": 0.7337,
+      "step": 1040
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0003263374485596708,
+      "loss": 0.7566,
+      "step": 1045
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0003259259259259259,
+      "loss": 0.7377,
+      "step": 1050
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00032551440329218106,
+      "loss": 0.7856,
+      "step": 1055
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0003251028806584362,
+      "loss": 0.7981,
+      "step": 1060
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0003246913580246914,
+      "loss": 0.7505,
+      "step": 1065
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0003242798353909465,
+      "loss": 0.7549,
+      "step": 1070
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00032386831275720164,
+      "loss": 0.7306,
+      "step": 1075
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0003234567901234568,
+      "loss": 0.7343,
+      "step": 1080
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0003230452674897119,
+      "loss": 0.738,
+      "step": 1085
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00032263374485596707,
+      "loss": 0.7655,
+      "step": 1090
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0003222222222222222,
+      "loss": 0.7706,
+      "step": 1095
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0003218106995884774,
+      "loss": 0.7512,
+      "step": 1100
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00032139917695473254,
+      "loss": 0.7549,
+      "step": 1105
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0003209876543209877,
+      "loss": 0.7839,
+      "step": 1110
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0003205761316872428,
+      "loss": 0.7511,
+      "step": 1115
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00032016460905349796,
+      "loss": 0.769,
+      "step": 1120
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0003197530864197531,
+      "loss": 0.7685,
+      "step": 1125
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0003193415637860083,
+      "loss": 0.7489,
+      "step": 1130
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0003189300411522634,
+      "loss": 0.808,
+      "step": 1135
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00031851851851851854,
+      "loss": 0.7654,
+      "step": 1140
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0003181069958847737,
+      "loss": 0.7845,
+      "step": 1145
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00031769547325102886,
+      "loss": 0.7886,
+      "step": 1150
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00031728395061728396,
+      "loss": 0.7318,
+      "step": 1155
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0003168724279835391,
+      "loss": 0.7231,
+      "step": 1160
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0003164609053497943,
+      "loss": 0.7471,
+      "step": 1165
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0003160493827160494,
+      "loss": 0.7424,
+      "step": 1170
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00031563786008230454,
+      "loss": 0.7249,
+      "step": 1175
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0003152263374485597,
+      "loss": 0.7419,
+      "step": 1180
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00031481481481481486,
+      "loss": 0.7567,
+      "step": 1185
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00031440329218106997,
+      "loss": 0.7707,
+      "step": 1190
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0003139917695473251,
+      "loss": 0.7473,
+      "step": 1195
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0003135802469135803,
+      "loss": 0.7222,
+      "step": 1200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0003131687242798354,
+      "loss": 0.7383,
+      "step": 1205
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00031275720164609055,
+      "loss": 0.7712,
+      "step": 1210
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0003123456790123457,
+      "loss": 0.7602,
+      "step": 1215
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0003119341563786008,
+      "loss": 0.7741,
+      "step": 1220
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00031152263374485597,
+      "loss": 0.7428,
+      "step": 1225
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0003111111111111111,
+      "loss": 0.7605,
+      "step": 1230
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0003106995884773663,
+      "loss": 0.7255,
+      "step": 1235
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0003102880658436214,
+      "loss": 0.7468,
+      "step": 1240
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00030987654320987655,
+      "loss": 0.7961,
+      "step": 1245
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0003094650205761317,
+      "loss": 0.7208,
+      "step": 1250
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0003090534979423868,
+      "loss": 0.7331,
+      "step": 1255
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00030864197530864197,
+      "loss": 0.7371,
+      "step": 1260
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00030823045267489713,
+      "loss": 0.7206,
+      "step": 1265
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0003078189300411523,
+      "loss": 0.7617,
+      "step": 1270
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0003074074074074074,
+      "loss": 0.7652,
+      "step": 1275
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00030699588477366255,
+      "loss": 0.7635,
+      "step": 1280
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0003065843621399177,
+      "loss": 0.7445,
+      "step": 1285
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0003061728395061728,
+      "loss": 0.7717,
+      "step": 1290
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00030576131687242797,
+      "loss": 0.7535,
+      "step": 1295
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00030534979423868313,
+      "loss": 0.7677,
+      "step": 1300
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0003049382716049383,
+      "loss": 0.8131,
+      "step": 1305
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00030452674897119345,
+      "loss": 0.7754,
+      "step": 1310
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00030411522633744855,
+      "loss": 0.76,
+      "step": 1315
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0003037037037037037,
+      "loss": 0.7558,
+      "step": 1320
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00030329218106995887,
+      "loss": 0.7604,
+      "step": 1325
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.000302880658436214,
+      "loss": 0.7716,
+      "step": 1330
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0003024691358024692,
+      "loss": 0.7669,
+      "step": 1335
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0003020576131687243,
+      "loss": 0.7469,
+      "step": 1340
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00030164609053497945,
+      "loss": 0.7511,
+      "step": 1345
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0003012345679012346,
+      "loss": 0.7483,
+      "step": 1350
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00030082304526748976,
+      "loss": 0.7509,
+      "step": 1355
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00030041152263374487,
+      "loss": 0.7249,
+      "step": 1360
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00030000000000000003,
+      "loss": 0.7289,
+      "step": 1365
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002995884773662552,
+      "loss": 0.7605,
+      "step": 1370
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002991769547325103,
+      "loss": 0.7185,
+      "step": 1375
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00029876543209876545,
+      "loss": 0.6974,
+      "step": 1380
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002983539094650206,
+      "loss": 0.7234,
+      "step": 1385
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00029794238683127577,
+      "loss": 0.7506,
+      "step": 1390
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00029753086419753087,
+      "loss": 0.7524,
+      "step": 1395
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00029711934156378603,
+      "loss": 0.7641,
+      "step": 1400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002967078189300412,
+      "loss": 0.6967,
+      "step": 1405
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002962962962962963,
+      "loss": 0.735,
+      "step": 1410
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00029588477366255145,
+      "loss": 0.7685,
+      "step": 1415
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002954732510288066,
+      "loss": 0.758,
+      "step": 1420
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002950617283950617,
+      "loss": 0.763,
+      "step": 1425
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00029465020576131687,
+      "loss": 0.712,
+      "step": 1430
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00029423868312757203,
+      "loss": 0.7236,
+      "step": 1435
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002938271604938272,
+      "loss": 0.7732,
+      "step": 1440
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002934156378600823,
+      "loss": 0.7689,
+      "step": 1445
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00029300411522633745,
+      "loss": 0.7352,
+      "step": 1450
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002925925925925926,
+      "loss": 0.7711,
+      "step": 1455
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002921810699588477,
+      "loss": 0.7549,
+      "step": 1460
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002917695473251029,
+      "loss": 0.7804,
+      "step": 1465
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00029135802469135803,
+      "loss": 0.7478,
+      "step": 1470
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002909465020576132,
+      "loss": 0.7801,
+      "step": 1475
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002905349794238683,
+      "loss": 0.7544,
+      "step": 1480
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00029012345679012345,
+      "loss": 0.7546,
+      "step": 1485
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002897119341563786,
+      "loss": 0.7648,
+      "step": 1490
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002893004115226337,
+      "loss": 0.7468,
+      "step": 1495
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002888888888888889,
+      "loss": 0.7434,
+      "step": 1500
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 5010,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 500,
+  "total_flos": 6.041910864582279e+18,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

save_lora_instruct_r8_4e4_s11_1/checkpoint-1500/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43458f4cf6aeb211c730b848f93ab15f483827c314fb4dc1e79dd1f658b7dbc8
+size 5944

save_lora_instruct_r8_4e4_s11_1/checkpoint-1500/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_4e4_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_4e4_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_4e4_s11_1/checkpoint-2000/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9671d2c504106bbfaae3fa086e35ace2d4b25fbf85e5c7020f54b5677be0e97e
+size 564342240

save_lora_instruct_r8_4e4_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 sha256:3933686a0f0a60d101b1032c5bef91c6fceaa0fe41c7f40320e09c7faa2cec0c
+size 239914224

save_lora_instruct_r8_4e4_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:a4e1ac50aa58a138b91832747cb96f8f87afd194acf2a86a3da4b96105025c18
+size 40303404

save_lora_instruct_r8_4e4_s11_1/checkpoint-2000/latest

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

save_lora_instruct_r8_4e4_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_4e4_s11_1/checkpoint-2000/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26d324026b4b66cd9b8553682d39e90a564b56f31b1a487c65e1e5128f325560
+size 1064

save_lora_instruct_r8_4e4_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": 1.3333333333333333e-05,
+      "loss": 1.3798,
+      "step": 5
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 2.6666666666666667e-05,
+      "loss": 1.3917,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 4e-05,
+      "loss": 1.2862,
+      "step": 15
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 5.333333333333333e-05,
+      "loss": 1.2046,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 6.666666666666667e-05,
+      "loss": 1.0634,
+      "step": 25
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 8e-05,
+      "loss": 1.0273,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.333333333333334e-05,
+      "loss": 1.0472,
+      "step": 35
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00010666666666666667,
+      "loss": 1.0301,
+      "step": 40
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00012,
+      "loss": 0.9335,
+      "step": 45
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00013333333333333334,
+      "loss": 0.921,
+      "step": 50
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00014666666666666666,
+      "loss": 0.9506,
+      "step": 55
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00016,
+      "loss": 0.8876,
+      "step": 60
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00017333333333333334,
+      "loss": 0.9335,
+      "step": 65
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0001866666666666667,
+      "loss": 0.8554,
+      "step": 70
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0002,
+      "loss": 0.8537,
+      "step": 75
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00021333333333333333,
+      "loss": 0.9033,
+      "step": 80
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00022666666666666668,
+      "loss": 0.8203,
+      "step": 85
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00024,
+      "loss": 0.8611,
+      "step": 90
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00025333333333333333,
+      "loss": 0.8303,
+      "step": 95
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0002666666666666667,
+      "loss": 0.8528,
+      "step": 100
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00028,
+      "loss": 0.8537,
+      "step": 105
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0002933333333333333,
+      "loss": 0.8286,
+      "step": 110
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0003066666666666667,
+      "loss": 0.8449,
+      "step": 115
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00032,
+      "loss": 0.8377,
+      "step": 120
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0003333333333333334,
+      "loss": 0.8419,
+      "step": 125
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00034666666666666667,
+      "loss": 0.8265,
+      "step": 130
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00036,
+      "loss": 0.8037,
+      "step": 135
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0003733333333333334,
+      "loss": 0.7924,
+      "step": 140
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00038666666666666667,
+      "loss": 0.8245,
+      "step": 145
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004,
+      "loss": 0.8388,
+      "step": 150
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0003995884773662552,
+      "loss": 0.8066,
+      "step": 155
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0003991769547325103,
+      "loss": 0.8107,
+      "step": 160
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00039876543209876544,
+      "loss": 0.8005,
+      "step": 165
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0003983539094650206,
+      "loss": 0.7727,
+      "step": 170
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00039794238683127576,
+      "loss": 0.8276,
+      "step": 175
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00039753086419753086,
+      "loss": 0.8464,
+      "step": 180
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000397119341563786,
+      "loss": 0.8348,
+      "step": 185
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0003967078189300412,
+      "loss": 0.8228,
+      "step": 190
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0003962962962962963,
+      "loss": 0.8254,
+      "step": 195
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00039588477366255144,
+      "loss": 0.817,
+      "step": 200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0003954732510288066,
+      "loss": 0.8365,
+      "step": 205
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0003950617283950617,
+      "loss": 0.793,
+      "step": 210
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00039465020576131686,
+      "loss": 0.8014,
+      "step": 215
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000394238683127572,
+      "loss": 0.8049,
+      "step": 220
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0003938271604938272,
+      "loss": 0.7841,
+      "step": 225
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0003934156378600823,
+      "loss": 0.7977,
+      "step": 230
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00039300411522633744,
+      "loss": 0.8299,
+      "step": 235
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0003925925925925926,
+      "loss": 0.7729,
+      "step": 240
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00039218106995884776,
+      "loss": 0.8082,
+      "step": 245
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0003917695473251029,
+      "loss": 0.8173,
+      "step": 250
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000391358024691358,
+      "loss": 0.8379,
+      "step": 255
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0003909465020576132,
+      "loss": 0.7962,
+      "step": 260
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00039053497942386834,
+      "loss": 0.8067,
+      "step": 265
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0003901234567901235,
+      "loss": 0.7915,
+      "step": 270
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00038971193415637866,
+      "loss": 0.7905,
+      "step": 275
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00038930041152263376,
+      "loss": 0.7862,
+      "step": 280
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0003888888888888889,
+      "loss": 0.7709,
+      "step": 285
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0003884773662551441,
+      "loss": 0.7773,
+      "step": 290
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00038806584362139924,
+      "loss": 0.8068,
+      "step": 295
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00038765432098765434,
+      "loss": 0.7962,
+      "step": 300
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0003872427983539095,
+      "loss": 0.7894,
+      "step": 305
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00038683127572016466,
+      "loss": 0.7705,
+      "step": 310
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00038641975308641976,
+      "loss": 0.7882,
+      "step": 315
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0003860082304526749,
+      "loss": 0.7865,
+      "step": 320
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0003855967078189301,
+      "loss": 0.7939,
+      "step": 325
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0003851851851851852,
+      "loss": 0.7724,
+      "step": 330
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00038477366255144034,
+      "loss": 0.7771,
+      "step": 335
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0003843621399176955,
+      "loss": 0.7941,
+      "step": 340
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00038395061728395066,
+      "loss": 0.7822,
+      "step": 345
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00038353909465020577,
+      "loss": 0.8143,
+      "step": 350
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0003831275720164609,
+      "loss": 0.795,
+      "step": 355
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0003827160493827161,
+      "loss": 0.8049,
+      "step": 360
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0003823045267489712,
+      "loss": 0.8309,
+      "step": 365
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00038189300411522635,
+      "loss": 0.8314,
+      "step": 370
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0003814814814814815,
+      "loss": 0.801,
+      "step": 375
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00038106995884773666,
+      "loss": 0.8084,
+      "step": 380
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00038065843621399177,
+      "loss": 0.8129,
+      "step": 385
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0003802469135802469,
+      "loss": 0.7859,
+      "step": 390
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0003798353909465021,
+      "loss": 0.7874,
+      "step": 395
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0003794238683127572,
+      "loss": 0.7548,
+      "step": 400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00037901234567901235,
+      "loss": 0.8092,
+      "step": 405
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0003786008230452675,
+      "loss": 0.8041,
+      "step": 410
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00037818930041152266,
+      "loss": 0.777,
+      "step": 415
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00037777777777777777,
+      "loss": 0.7699,
+      "step": 420
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00037736625514403293,
+      "loss": 0.781,
+      "step": 425
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0003769547325102881,
+      "loss": 0.7504,
+      "step": 430
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0003765432098765432,
+      "loss": 0.7974,
+      "step": 435
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00037613168724279835,
+      "loss": 0.7663,
+      "step": 440
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0003757201646090535,
+      "loss": 0.8013,
+      "step": 445
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00037530864197530867,
+      "loss": 0.7453,
+      "step": 450
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00037489711934156377,
+      "loss": 0.7425,
+      "step": 455
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00037448559670781893,
+      "loss": 0.7582,
+      "step": 460
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0003740740740740741,
+      "loss": 0.7801,
+      "step": 465
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00037366255144032925,
+      "loss": 0.7809,
+      "step": 470
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0003732510288065844,
+      "loss": 0.7749,
+      "step": 475
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00037283950617283956,
+      "loss": 0.7807,
+      "step": 480
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00037242798353909467,
+      "loss": 0.7715,
+      "step": 485
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0003720164609053498,
+      "loss": 0.7356,
+      "step": 490
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.000371604938271605,
+      "loss": 0.7674,
+      "step": 495
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00037119341563786014,
+      "loss": 0.7542,
+      "step": 500
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00037078189300411525,
+      "loss": 0.7583,
+      "step": 505
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0003703703703703704,
+      "loss": 0.7767,
+      "step": 510
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00036995884773662556,
+      "loss": 0.7853,
+      "step": 515
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00036954732510288067,
+      "loss": 0.7733,
+      "step": 520
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00036913580246913583,
+      "loss": 0.7758,
+      "step": 525
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.000368724279835391,
+      "loss": 0.7698,
+      "step": 530
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00036831275720164614,
+      "loss": 0.7805,
+      "step": 535
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00036790123456790125,
+      "loss": 0.7573,
+      "step": 540
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003674897119341564,
+      "loss": 0.7709,
+      "step": 545
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00036707818930041157,
+      "loss": 0.7994,
+      "step": 550
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00036666666666666667,
+      "loss": 0.7363,
+      "step": 555
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00036625514403292183,
+      "loss": 0.7478,
+      "step": 560
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.000365843621399177,
+      "loss": 0.8188,
+      "step": 565
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003654320987654321,
+      "loss": 0.7524,
+      "step": 570
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00036502057613168725,
+      "loss": 0.7319,
+      "step": 575
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003646090534979424,
+      "loss": 0.7915,
+      "step": 580
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00036419753086419757,
+      "loss": 0.7722,
+      "step": 585
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00036378600823045267,
+      "loss": 0.7842,
+      "step": 590
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00036337448559670783,
+      "loss": 0.7591,
+      "step": 595
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.000362962962962963,
+      "loss": 0.7653,
+      "step": 600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003625514403292181,
+      "loss": 0.7838,
+      "step": 605
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00036213991769547325,
+      "loss": 0.7937,
+      "step": 610
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003617283950617284,
+      "loss": 0.77,
+      "step": 615
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00036131687242798357,
+      "loss": 0.7899,
+      "step": 620
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003609053497942387,
+      "loss": 0.7815,
+      "step": 625
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00036049382716049383,
+      "loss": 0.7648,
+      "step": 630
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.000360082304526749,
+      "loss": 0.7716,
+      "step": 635
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003596707818930041,
+      "loss": 0.7708,
+      "step": 640
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00035925925925925925,
+      "loss": 0.7306,
+      "step": 645
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003588477366255144,
+      "loss": 0.8007,
+      "step": 650
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003584362139917695,
+      "loss": 0.7551,
+      "step": 655
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003580246913580247,
+      "loss": 0.754,
+      "step": 660
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00035761316872427983,
+      "loss": 0.7599,
+      "step": 665
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.000357201646090535,
+      "loss": 0.7637,
+      "step": 670
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00035679012345679015,
+      "loss": 0.7766,
+      "step": 675
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003563786008230453,
+      "loss": 0.7727,
+      "step": 680
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003559670781893004,
+      "loss": 0.7675,
+      "step": 685
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00035555555555555557,
+      "loss": 0.7707,
+      "step": 690
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00035514403292181073,
+      "loss": 0.7221,
+      "step": 695
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003547325102880659,
+      "loss": 0.7335,
+      "step": 700
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00035432098765432105,
+      "loss": 0.781,
+      "step": 705
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00035390946502057615,
+      "loss": 0.7845,
+      "step": 710
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003534979423868313,
+      "loss": 0.7467,
+      "step": 715
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035308641975308647,
+      "loss": 0.7735,
+      "step": 720
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003526748971193416,
+      "loss": 0.7368,
+      "step": 725
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035226337448559673,
+      "loss": 0.7678,
+      "step": 730
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003518518518518519,
+      "loss": 0.7756,
+      "step": 735
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035144032921810705,
+      "loss": 0.7635,
+      "step": 740
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035102880658436215,
+      "loss": 0.7757,
+      "step": 745
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003506172839506173,
+      "loss": 0.7442,
+      "step": 750
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035020576131687247,
+      "loss": 0.7584,
+      "step": 755
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003497942386831276,
+      "loss": 0.8333,
+      "step": 760
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034938271604938273,
+      "loss": 0.7459,
+      "step": 765
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003489711934156379,
+      "loss": 0.7499,
+      "step": 770
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.000348559670781893,
+      "loss": 0.7571,
+      "step": 775
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034814814814814816,
+      "loss": 0.7243,
+      "step": 780
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003477366255144033,
+      "loss": 0.7443,
+      "step": 785
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003473251028806585,
+      "loss": 0.7759,
+      "step": 790
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003469135802469136,
+      "loss": 0.7452,
+      "step": 795
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034650205761316874,
+      "loss": 0.7578,
+      "step": 800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003460905349794239,
+      "loss": 0.7634,
+      "step": 805
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.000345679012345679,
+      "loss": 0.7925,
+      "step": 810
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00034526748971193416,
+      "loss": 0.7966,
+      "step": 815
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003448559670781893,
+      "loss": 0.7601,
+      "step": 820
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003444444444444445,
+      "loss": 0.7617,
+      "step": 825
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003440329218106996,
+      "loss": 0.7526,
+      "step": 830
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00034362139917695474,
+      "loss": 0.7297,
+      "step": 835
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003432098765432099,
+      "loss": 0.7549,
+      "step": 840
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.000342798353909465,
+      "loss": 0.7517,
+      "step": 845
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00034238683127572016,
+      "loss": 0.7626,
+      "step": 850
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003419753086419753,
+      "loss": 0.7987,
+      "step": 855
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003415637860082305,
+      "loss": 0.7573,
+      "step": 860
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003411522633744856,
+      "loss": 0.8367,
+      "step": 865
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00034074074074074074,
+      "loss": 0.7258,
+      "step": 870
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003403292181069959,
+      "loss": 0.7632,
+      "step": 875
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00033991769547325106,
+      "loss": 0.7379,
+      "step": 880
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00033950617283950616,
+      "loss": 0.7529,
+      "step": 885
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003390946502057613,
+      "loss": 0.7391,
+      "step": 890
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003386831275720165,
+      "loss": 0.8061,
+      "step": 895
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00033827160493827164,
+      "loss": 0.7288,
+      "step": 900
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003378600823045268,
+      "loss": 0.7493,
+      "step": 905
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00033744855967078195,
+      "loss": 0.7702,
+      "step": 910
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00033703703703703706,
+      "loss": 0.7538,
+      "step": 915
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003366255144032922,
+      "loss": 0.728,
+      "step": 920
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003362139917695474,
+      "loss": 0.7776,
+      "step": 925
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003358024691358025,
+      "loss": 0.7883,
+      "step": 930
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00033539094650205764,
+      "loss": 0.7471,
+      "step": 935
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003349794238683128,
+      "loss": 0.7464,
+      "step": 940
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00033456790123456795,
+      "loss": 0.7885,
+      "step": 945
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00033415637860082306,
+      "loss": 0.7509,
+      "step": 950
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003337448559670782,
+      "loss": 0.7681,
+      "step": 955
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003333333333333334,
+      "loss": 0.8246,
+      "step": 960
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003329218106995885,
+      "loss": 0.7604,
+      "step": 965
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00033251028806584364,
+      "loss": 0.7755,
+      "step": 970
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003320987654320988,
+      "loss": 0.7641,
+      "step": 975
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003316872427983539,
+      "loss": 0.7708,
+      "step": 980
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00033127572016460906,
+      "loss": 0.7936,
+      "step": 985
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003308641975308642,
+      "loss": 0.7641,
+      "step": 990
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003304526748971194,
+      "loss": 0.7479,
+      "step": 995
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003300411522633745,
+      "loss": 0.7547,
+      "step": 1000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00032962962962962964,
+      "loss": 0.762,
+      "step": 1005
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003292181069958848,
+      "loss": 0.7472,
+      "step": 1010
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003288065843621399,
+      "loss": 0.7542,
+      "step": 1015
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00032839506172839506,
+      "loss": 0.7669,
+      "step": 1020
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003279835390946502,
+      "loss": 0.7812,
+      "step": 1025
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003275720164609054,
+      "loss": 0.7273,
+      "step": 1030
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003271604938271605,
+      "loss": 0.7822,
+      "step": 1035
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00032674897119341564,
+      "loss": 0.7337,
+      "step": 1040
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0003263374485596708,
+      "loss": 0.7566,
+      "step": 1045
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0003259259259259259,
+      "loss": 0.7377,
+      "step": 1050
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00032551440329218106,
+      "loss": 0.7856,
+      "step": 1055
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0003251028806584362,
+      "loss": 0.7981,
+      "step": 1060
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0003246913580246914,
+      "loss": 0.7505,
+      "step": 1065
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0003242798353909465,
+      "loss": 0.7549,
+      "step": 1070
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00032386831275720164,
+      "loss": 0.7306,
+      "step": 1075
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0003234567901234568,
+      "loss": 0.7343,
+      "step": 1080
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0003230452674897119,
+      "loss": 0.738,
+      "step": 1085
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00032263374485596707,
+      "loss": 0.7655,
+      "step": 1090
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0003222222222222222,
+      "loss": 0.7706,
+      "step": 1095
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0003218106995884774,
+      "loss": 0.7512,
+      "step": 1100
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00032139917695473254,
+      "loss": 0.7549,
+      "step": 1105
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0003209876543209877,
+      "loss": 0.7839,
+      "step": 1110
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0003205761316872428,
+      "loss": 0.7511,
+      "step": 1115
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00032016460905349796,
+      "loss": 0.769,
+      "step": 1120
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0003197530864197531,
+      "loss": 0.7685,
+      "step": 1125
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0003193415637860083,
+      "loss": 0.7489,
+      "step": 1130
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0003189300411522634,
+      "loss": 0.808,
+      "step": 1135
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00031851851851851854,
+      "loss": 0.7654,
+      "step": 1140
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0003181069958847737,
+      "loss": 0.7845,
+      "step": 1145
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00031769547325102886,
+      "loss": 0.7886,
+      "step": 1150
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00031728395061728396,
+      "loss": 0.7318,
+      "step": 1155
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0003168724279835391,
+      "loss": 0.7231,
+      "step": 1160
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0003164609053497943,
+      "loss": 0.7471,
+      "step": 1165
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0003160493827160494,
+      "loss": 0.7424,
+      "step": 1170
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00031563786008230454,
+      "loss": 0.7249,
+      "step": 1175
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0003152263374485597,
+      "loss": 0.7419,
+      "step": 1180
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00031481481481481486,
+      "loss": 0.7567,
+      "step": 1185
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00031440329218106997,
+      "loss": 0.7707,
+      "step": 1190
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0003139917695473251,
+      "loss": 0.7473,
+      "step": 1195
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0003135802469135803,
+      "loss": 0.7222,
+      "step": 1200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0003131687242798354,
+      "loss": 0.7383,
+      "step": 1205
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00031275720164609055,
+      "loss": 0.7712,
+      "step": 1210
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0003123456790123457,
+      "loss": 0.7602,
+      "step": 1215
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0003119341563786008,
+      "loss": 0.7741,
+      "step": 1220
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00031152263374485597,
+      "loss": 0.7428,
+      "step": 1225
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0003111111111111111,
+      "loss": 0.7605,
+      "step": 1230
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0003106995884773663,
+      "loss": 0.7255,
+      "step": 1235
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0003102880658436214,
+      "loss": 0.7468,
+      "step": 1240
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00030987654320987655,
+      "loss": 0.7961,
+      "step": 1245
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0003094650205761317,
+      "loss": 0.7208,
+      "step": 1250
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0003090534979423868,
+      "loss": 0.7331,
+      "step": 1255
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00030864197530864197,
+      "loss": 0.7371,
+      "step": 1260
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00030823045267489713,
+      "loss": 0.7206,
+      "step": 1265
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0003078189300411523,
+      "loss": 0.7617,
+      "step": 1270
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0003074074074074074,
+      "loss": 0.7652,
+      "step": 1275
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00030699588477366255,
+      "loss": 0.7635,
+      "step": 1280
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0003065843621399177,
+      "loss": 0.7445,
+      "step": 1285
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0003061728395061728,
+      "loss": 0.7717,
+      "step": 1290
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00030576131687242797,
+      "loss": 0.7535,
+      "step": 1295
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00030534979423868313,
+      "loss": 0.7677,
+      "step": 1300
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0003049382716049383,
+      "loss": 0.8131,
+      "step": 1305
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00030452674897119345,
+      "loss": 0.7754,
+      "step": 1310
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00030411522633744855,
+      "loss": 0.76,
+      "step": 1315
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0003037037037037037,
+      "loss": 0.7558,
+      "step": 1320
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00030329218106995887,
+      "loss": 0.7604,
+      "step": 1325
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.000302880658436214,
+      "loss": 0.7716,
+      "step": 1330
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0003024691358024692,
+      "loss": 0.7669,
+      "step": 1335
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0003020576131687243,
+      "loss": 0.7469,
+      "step": 1340
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00030164609053497945,
+      "loss": 0.7511,
+      "step": 1345
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0003012345679012346,
+      "loss": 0.7483,
+      "step": 1350
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00030082304526748976,
+      "loss": 0.7509,
+      "step": 1355
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00030041152263374487,
+      "loss": 0.7249,
+      "step": 1360
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00030000000000000003,
+      "loss": 0.7289,
+      "step": 1365
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002995884773662552,
+      "loss": 0.7605,
+      "step": 1370
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002991769547325103,
+      "loss": 0.7185,
+      "step": 1375
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00029876543209876545,
+      "loss": 0.6974,
+      "step": 1380
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002983539094650206,
+      "loss": 0.7234,
+      "step": 1385
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00029794238683127577,
+      "loss": 0.7506,
+      "step": 1390
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00029753086419753087,
+      "loss": 0.7524,
+      "step": 1395
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00029711934156378603,
+      "loss": 0.7641,
+      "step": 1400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002967078189300412,
+      "loss": 0.6967,
+      "step": 1405
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002962962962962963,
+      "loss": 0.735,
+      "step": 1410
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00029588477366255145,
+      "loss": 0.7685,
+      "step": 1415
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002954732510288066,
+      "loss": 0.758,
+      "step": 1420
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002950617283950617,
+      "loss": 0.763,
+      "step": 1425
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00029465020576131687,
+      "loss": 0.712,
+      "step": 1430
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00029423868312757203,
+      "loss": 0.7236,
+      "step": 1435
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002938271604938272,
+      "loss": 0.7732,
+      "step": 1440
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002934156378600823,
+      "loss": 0.7689,
+      "step": 1445
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00029300411522633745,
+      "loss": 0.7352,
+      "step": 1450
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002925925925925926,
+      "loss": 0.7711,
+      "step": 1455
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002921810699588477,
+      "loss": 0.7549,
+      "step": 1460
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002917695473251029,
+      "loss": 0.7804,
+      "step": 1465
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00029135802469135803,
+      "loss": 0.7478,
+      "step": 1470
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002909465020576132,
+      "loss": 0.7801,
+      "step": 1475
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002905349794238683,
+      "loss": 0.7544,
+      "step": 1480
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00029012345679012345,
+      "loss": 0.7546,
+      "step": 1485
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002897119341563786,
+      "loss": 0.7648,
+      "step": 1490
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002893004115226337,
+      "loss": 0.7468,
+      "step": 1495
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002888888888888889,
+      "loss": 0.7434,
+      "step": 1500
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00028847736625514403,
+      "loss": 0.7348,
+      "step": 1505
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002880658436213992,
+      "loss": 0.7515,
+      "step": 1510
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00028765432098765435,
+      "loss": 0.7345,
+      "step": 1515
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00028724279835390946,
+      "loss": 0.7701,
+      "step": 1520
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0002868312757201646,
+      "loss": 0.7472,
+      "step": 1525
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00028641975308641977,
+      "loss": 0.7075,
+      "step": 1530
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00028600823045267493,
+      "loss": 0.7483,
+      "step": 1535
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0002855967078189301,
+      "loss": 0.712,
+      "step": 1540
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002851851851851852,
+      "loss": 0.7484,
+      "step": 1545
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00028477366255144035,
+      "loss": 0.7471,
+      "step": 1550
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002843621399176955,
+      "loss": 0.7462,
+      "step": 1555
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00028395061728395067,
+      "loss": 0.7371,
+      "step": 1560
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002835390946502058,
+      "loss": 0.7753,
+      "step": 1565
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00028312757201646093,
+      "loss": 0.7374,
+      "step": 1570
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002827160493827161,
+      "loss": 0.7328,
+      "step": 1575
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002823045267489712,
+      "loss": 0.7159,
+      "step": 1580
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00028189300411522635,
+      "loss": 0.7594,
+      "step": 1585
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002814814814814815,
+      "loss": 0.7215,
+      "step": 1590
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00028106995884773667,
+      "loss": 0.7398,
+      "step": 1595
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0002806584362139918,
+      "loss": 0.773,
+      "step": 1600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00028024691358024693,
+      "loss": 0.7584,
+      "step": 1605
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0002798353909465021,
+      "loss": 0.7117,
+      "step": 1610
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0002794238683127572,
+      "loss": 0.7471,
+      "step": 1615
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00027901234567901236,
+      "loss": 0.7443,
+      "step": 1620
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002786008230452675,
+      "loss": 0.7292,
+      "step": 1625
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002781893004115227,
+      "loss": 0.7716,
+      "step": 1630
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002777777777777778,
+      "loss": 0.763,
+      "step": 1635
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00027736625514403294,
+      "loss": 0.7408,
+      "step": 1640
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002769547325102881,
+      "loss": 0.7403,
+      "step": 1645
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002765432098765432,
+      "loss": 0.7508,
+      "step": 1650
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00027613168724279836,
+      "loss": 0.7687,
+      "step": 1655
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002757201646090535,
+      "loss": 0.7305,
+      "step": 1660
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002753086419753086,
+      "loss": 0.7165,
+      "step": 1665
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002748971193415638,
+      "loss": 0.8,
+      "step": 1670
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00027448559670781894,
+      "loss": 0.7357,
+      "step": 1675
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002740740740740741,
+      "loss": 0.7988,
+      "step": 1680
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002736625514403292,
+      "loss": 0.7559,
+      "step": 1685
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00027325102880658436,
+      "loss": 0.712,
+      "step": 1690
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002728395061728395,
+      "loss": 0.7486,
+      "step": 1695
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002724279835390946,
+      "loss": 0.7388,
+      "step": 1700
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002720164609053498,
+      "loss": 0.7441,
+      "step": 1705
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00027160493827160494,
+      "loss": 0.7547,
+      "step": 1710
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002711934156378601,
+      "loss": 0.7229,
+      "step": 1715
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002707818930041152,
+      "loss": 0.7383,
+      "step": 1720
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00027037037037037036,
+      "loss": 0.7606,
+      "step": 1725
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002699588477366255,
+      "loss": 0.7529,
+      "step": 1730
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002695473251028807,
+      "loss": 0.752,
+      "step": 1735
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00026913580246913584,
+      "loss": 0.7342,
+      "step": 1740
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.000268724279835391,
+      "loss": 0.7308,
+      "step": 1745
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002683127572016461,
+      "loss": 0.6915,
+      "step": 1750
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00026790123456790126,
+      "loss": 0.7344,
+      "step": 1755
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002674897119341564,
+      "loss": 0.7426,
+      "step": 1760
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002670781893004116,
+      "loss": 0.7088,
+      "step": 1765
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0002666666666666667,
+      "loss": 0.7241,
+      "step": 1770
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00026625514403292184,
+      "loss": 0.7552,
+      "step": 1775
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.000265843621399177,
+      "loss": 0.7477,
+      "step": 1780
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0002654320987654321,
+      "loss": 0.7115,
+      "step": 1785
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00026502057613168726,
+      "loss": 0.7538,
+      "step": 1790
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0002646090534979424,
+      "loss": 0.7321,
+      "step": 1795
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0002641975308641976,
+      "loss": 0.7212,
+      "step": 1800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0002637860082304527,
+      "loss": 0.747,
+      "step": 1805
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00026337448559670784,
+      "loss": 0.7606,
+      "step": 1810
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.000262962962962963,
+      "loss": 0.7344,
+      "step": 1815
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002625514403292181,
+      "loss": 0.7783,
+      "step": 1820
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00026213991769547326,
+      "loss": 0.7556,
+      "step": 1825
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002617283950617284,
+      "loss": 0.7585,
+      "step": 1830
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002613168724279836,
+      "loss": 0.7359,
+      "step": 1835
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002609053497942387,
+      "loss": 0.7185,
+      "step": 1840
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00026049382716049384,
+      "loss": 0.7573,
+      "step": 1845
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.000260082304526749,
+      "loss": 0.7619,
+      "step": 1850
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0002596707818930041,
+      "loss": 0.7611,
+      "step": 1855
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00025925925925925926,
+      "loss": 0.7647,
+      "step": 1860
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0002588477366255144,
+      "loss": 0.7398,
+      "step": 1865
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0002584362139917695,
+      "loss": 0.7683,
+      "step": 1870
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0002580246913580247,
+      "loss": 0.7817,
+      "step": 1875
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00025761316872427984,
+      "loss": 0.742,
+      "step": 1880
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.000257201646090535,
+      "loss": 0.7509,
+      "step": 1885
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0002567901234567901,
+      "loss": 0.7058,
+      "step": 1890
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00025637860082304526,
+      "loss": 0.7635,
+      "step": 1895
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0002559670781893004,
+      "loss": 0.734,
+      "step": 1900
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00025555555555555553,
+      "loss": 0.7416,
+      "step": 1905
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0002551440329218107,
+      "loss": 0.7071,
+      "step": 1910
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00025473251028806584,
+      "loss": 0.7275,
+      "step": 1915
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.000254320987654321,
+      "loss": 0.7413,
+      "step": 1920
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0002539094650205761,
+      "loss": 0.7703,
+      "step": 1925
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00025349794238683127,
+      "loss": 0.7331,
+      "step": 1930
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0002530864197530864,
+      "loss": 0.7251,
+      "step": 1935
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0002526748971193416,
+      "loss": 0.7341,
+      "step": 1940
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00025226337448559674,
+      "loss": 0.7079,
+      "step": 1945
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00025185185185185185,
+      "loss": 0.7615,
+      "step": 1950
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.000251440329218107,
+      "loss": 0.7144,
+      "step": 1955
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00025102880658436216,
+      "loss": 0.7527,
+      "step": 1960
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0002506172839506173,
+      "loss": 0.7327,
+      "step": 1965
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0002502057613168725,
+      "loss": 0.7173,
+      "step": 1970
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0002497942386831276,
+      "loss": 0.7239,
+      "step": 1975
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00024938271604938274,
+      "loss": 0.7442,
+      "step": 1980
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0002489711934156379,
+      "loss": 0.7396,
+      "step": 1985
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.000248559670781893,
+      "loss": 0.7365,
+      "step": 1990
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00024814814814814816,
+      "loss": 0.7037,
+      "step": 1995
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0002477366255144033,
+      "loss": 0.721,
+      "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_4e4_s11_1/checkpoint-2000/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43458f4cf6aeb211c730b848f93ab15f483827c314fb4dc1e79dd1f658b7dbc8
+size 5944

save_lora_instruct_r8_4e4_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_4e4_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_4e4_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_4e4_s11_1/checkpoint-2500/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b416d3e258a582396075c05428bff40bfdfca1eb90c49c8d3772f778275bc31
+size 564342240

save_lora_instruct_r8_4e4_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:dcdc9e140d08d1a6a769adf0c31faed7eb8110dc4b7f6518a08d49a23e02294f
+size 239914224

save_lora_instruct_r8_4e4_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:f60f73c4622f20dc7ff3cfbb9b219fee35bf8134a14b777c6845e321c049fb20
+size 40303404

save_lora_instruct_r8_4e4_s11_1/checkpoint-2500/latest

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

save_lora_instruct_r8_4e4_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_4e4_s11_1/checkpoint-2500/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a13f4f8ba2e5988cc6fd82905cf4ea363a21a822e658eabdc1f337bcaed20957
+size 1064

save_lora_instruct_r8_4e4_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": 1.3333333333333333e-05,
+      "loss": 1.3798,
+      "step": 5
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 2.6666666666666667e-05,
+      "loss": 1.3917,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 4e-05,
+      "loss": 1.2862,
+      "step": 15
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 5.333333333333333e-05,
+      "loss": 1.2046,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 6.666666666666667e-05,
+      "loss": 1.0634,
+      "step": 25
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 8e-05,
+      "loss": 1.0273,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.333333333333334e-05,
+      "loss": 1.0472,
+      "step": 35
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00010666666666666667,
+      "loss": 1.0301,
+      "step": 40
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00012,
+      "loss": 0.9335,
+      "step": 45
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00013333333333333334,
+      "loss": 0.921,
+      "step": 50
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00014666666666666666,
+      "loss": 0.9506,
+      "step": 55
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00016,
+      "loss": 0.8876,
+      "step": 60
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00017333333333333334,
+      "loss": 0.9335,
+      "step": 65
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0001866666666666667,
+      "loss": 0.8554,
+      "step": 70
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0002,
+      "loss": 0.8537,
+      "step": 75
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00021333333333333333,
+      "loss": 0.9033,
+      "step": 80
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00022666666666666668,
+      "loss": 0.8203,
+      "step": 85
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00024,
+      "loss": 0.8611,
+      "step": 90
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00025333333333333333,
+      "loss": 0.8303,
+      "step": 95
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0002666666666666667,
+      "loss": 0.8528,
+      "step": 100
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00028,
+      "loss": 0.8537,
+      "step": 105
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0002933333333333333,
+      "loss": 0.8286,
+      "step": 110
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0003066666666666667,
+      "loss": 0.8449,
+      "step": 115
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00032,
+      "loss": 0.8377,
+      "step": 120
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0003333333333333334,
+      "loss": 0.8419,
+      "step": 125
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00034666666666666667,
+      "loss": 0.8265,
+      "step": 130
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00036,
+      "loss": 0.8037,
+      "step": 135
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0003733333333333334,
+      "loss": 0.7924,
+      "step": 140
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00038666666666666667,
+      "loss": 0.8245,
+      "step": 145
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004,
+      "loss": 0.8388,
+      "step": 150
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0003995884773662552,
+      "loss": 0.8066,
+      "step": 155
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0003991769547325103,
+      "loss": 0.8107,
+      "step": 160
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00039876543209876544,
+      "loss": 0.8005,
+      "step": 165
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0003983539094650206,
+      "loss": 0.7727,
+      "step": 170
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00039794238683127576,
+      "loss": 0.8276,
+      "step": 175
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00039753086419753086,
+      "loss": 0.8464,
+      "step": 180
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000397119341563786,
+      "loss": 0.8348,
+      "step": 185
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0003967078189300412,
+      "loss": 0.8228,
+      "step": 190
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0003962962962962963,
+      "loss": 0.8254,
+      "step": 195
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00039588477366255144,
+      "loss": 0.817,
+      "step": 200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0003954732510288066,
+      "loss": 0.8365,
+      "step": 205
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0003950617283950617,
+      "loss": 0.793,
+      "step": 210
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00039465020576131686,
+      "loss": 0.8014,
+      "step": 215
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000394238683127572,
+      "loss": 0.8049,
+      "step": 220
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0003938271604938272,
+      "loss": 0.7841,
+      "step": 225
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0003934156378600823,
+      "loss": 0.7977,
+      "step": 230
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00039300411522633744,
+      "loss": 0.8299,
+      "step": 235
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0003925925925925926,
+      "loss": 0.7729,
+      "step": 240
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00039218106995884776,
+      "loss": 0.8082,
+      "step": 245
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0003917695473251029,
+      "loss": 0.8173,
+      "step": 250
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000391358024691358,
+      "loss": 0.8379,
+      "step": 255
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0003909465020576132,
+      "loss": 0.7962,
+      "step": 260
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00039053497942386834,
+      "loss": 0.8067,
+      "step": 265
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0003901234567901235,
+      "loss": 0.7915,
+      "step": 270
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00038971193415637866,
+      "loss": 0.7905,
+      "step": 275
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00038930041152263376,
+      "loss": 0.7862,
+      "step": 280
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0003888888888888889,
+      "loss": 0.7709,
+      "step": 285
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0003884773662551441,
+      "loss": 0.7773,
+      "step": 290
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00038806584362139924,
+      "loss": 0.8068,
+      "step": 295
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00038765432098765434,
+      "loss": 0.7962,
+      "step": 300
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0003872427983539095,
+      "loss": 0.7894,
+      "step": 305
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00038683127572016466,
+      "loss": 0.7705,
+      "step": 310
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00038641975308641976,
+      "loss": 0.7882,
+      "step": 315
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0003860082304526749,
+      "loss": 0.7865,
+      "step": 320
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0003855967078189301,
+      "loss": 0.7939,
+      "step": 325
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0003851851851851852,
+      "loss": 0.7724,
+      "step": 330
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00038477366255144034,
+      "loss": 0.7771,
+      "step": 335
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0003843621399176955,
+      "loss": 0.7941,
+      "step": 340
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00038395061728395066,
+      "loss": 0.7822,
+      "step": 345
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00038353909465020577,
+      "loss": 0.8143,
+      "step": 350
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0003831275720164609,
+      "loss": 0.795,
+      "step": 355
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0003827160493827161,
+      "loss": 0.8049,
+      "step": 360
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0003823045267489712,
+      "loss": 0.8309,
+      "step": 365
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00038189300411522635,
+      "loss": 0.8314,
+      "step": 370
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0003814814814814815,
+      "loss": 0.801,
+      "step": 375
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00038106995884773666,
+      "loss": 0.8084,
+      "step": 380
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00038065843621399177,
+      "loss": 0.8129,
+      "step": 385
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0003802469135802469,
+      "loss": 0.7859,
+      "step": 390
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0003798353909465021,
+      "loss": 0.7874,
+      "step": 395
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0003794238683127572,
+      "loss": 0.7548,
+      "step": 400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00037901234567901235,
+      "loss": 0.8092,
+      "step": 405
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0003786008230452675,
+      "loss": 0.8041,
+      "step": 410
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00037818930041152266,
+      "loss": 0.777,
+      "step": 415
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00037777777777777777,
+      "loss": 0.7699,
+      "step": 420
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00037736625514403293,
+      "loss": 0.781,
+      "step": 425
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0003769547325102881,
+      "loss": 0.7504,
+      "step": 430
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0003765432098765432,
+      "loss": 0.7974,
+      "step": 435
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00037613168724279835,
+      "loss": 0.7663,
+      "step": 440
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0003757201646090535,
+      "loss": 0.8013,
+      "step": 445
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00037530864197530867,
+      "loss": 0.7453,
+      "step": 450
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00037489711934156377,
+      "loss": 0.7425,
+      "step": 455
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00037448559670781893,
+      "loss": 0.7582,
+      "step": 460
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0003740740740740741,
+      "loss": 0.7801,
+      "step": 465
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00037366255144032925,
+      "loss": 0.7809,
+      "step": 470
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0003732510288065844,
+      "loss": 0.7749,
+      "step": 475
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00037283950617283956,
+      "loss": 0.7807,
+      "step": 480
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00037242798353909467,
+      "loss": 0.7715,
+      "step": 485
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0003720164609053498,
+      "loss": 0.7356,
+      "step": 490
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.000371604938271605,
+      "loss": 0.7674,
+      "step": 495
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00037119341563786014,
+      "loss": 0.7542,
+      "step": 500
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00037078189300411525,
+      "loss": 0.7583,
+      "step": 505
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0003703703703703704,
+      "loss": 0.7767,
+      "step": 510
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00036995884773662556,
+      "loss": 0.7853,
+      "step": 515
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00036954732510288067,
+      "loss": 0.7733,
+      "step": 520
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00036913580246913583,
+      "loss": 0.7758,
+      "step": 525
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.000368724279835391,
+      "loss": 0.7698,
+      "step": 530
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00036831275720164614,
+      "loss": 0.7805,
+      "step": 535
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00036790123456790125,
+      "loss": 0.7573,
+      "step": 540
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003674897119341564,
+      "loss": 0.7709,
+      "step": 545
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00036707818930041157,
+      "loss": 0.7994,
+      "step": 550
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00036666666666666667,
+      "loss": 0.7363,
+      "step": 555
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00036625514403292183,
+      "loss": 0.7478,
+      "step": 560
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.000365843621399177,
+      "loss": 0.8188,
+      "step": 565
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003654320987654321,
+      "loss": 0.7524,
+      "step": 570
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00036502057613168725,
+      "loss": 0.7319,
+      "step": 575
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003646090534979424,
+      "loss": 0.7915,
+      "step": 580
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00036419753086419757,
+      "loss": 0.7722,
+      "step": 585
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00036378600823045267,
+      "loss": 0.7842,
+      "step": 590
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00036337448559670783,
+      "loss": 0.7591,
+      "step": 595
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.000362962962962963,
+      "loss": 0.7653,
+      "step": 600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003625514403292181,
+      "loss": 0.7838,
+      "step": 605
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00036213991769547325,
+      "loss": 0.7937,
+      "step": 610
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003617283950617284,
+      "loss": 0.77,
+      "step": 615
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00036131687242798357,
+      "loss": 0.7899,
+      "step": 620
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003609053497942387,
+      "loss": 0.7815,
+      "step": 625
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00036049382716049383,
+      "loss": 0.7648,
+      "step": 630
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.000360082304526749,
+      "loss": 0.7716,
+      "step": 635
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003596707818930041,
+      "loss": 0.7708,
+      "step": 640
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00035925925925925925,
+      "loss": 0.7306,
+      "step": 645
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003588477366255144,
+      "loss": 0.8007,
+      "step": 650
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003584362139917695,
+      "loss": 0.7551,
+      "step": 655
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003580246913580247,
+      "loss": 0.754,
+      "step": 660
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00035761316872427983,
+      "loss": 0.7599,
+      "step": 665
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.000357201646090535,
+      "loss": 0.7637,
+      "step": 670
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00035679012345679015,
+      "loss": 0.7766,
+      "step": 675
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003563786008230453,
+      "loss": 0.7727,
+      "step": 680
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003559670781893004,
+      "loss": 0.7675,
+      "step": 685
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00035555555555555557,
+      "loss": 0.7707,
+      "step": 690
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00035514403292181073,
+      "loss": 0.7221,
+      "step": 695
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003547325102880659,
+      "loss": 0.7335,
+      "step": 700
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00035432098765432105,
+      "loss": 0.781,
+      "step": 705
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00035390946502057615,
+      "loss": 0.7845,
+      "step": 710
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003534979423868313,
+      "loss": 0.7467,
+      "step": 715
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035308641975308647,
+      "loss": 0.7735,
+      "step": 720
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003526748971193416,
+      "loss": 0.7368,
+      "step": 725
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035226337448559673,
+      "loss": 0.7678,
+      "step": 730
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003518518518518519,
+      "loss": 0.7756,
+      "step": 735
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035144032921810705,
+      "loss": 0.7635,
+      "step": 740
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035102880658436215,
+      "loss": 0.7757,
+      "step": 745
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003506172839506173,
+      "loss": 0.7442,
+      "step": 750
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035020576131687247,
+      "loss": 0.7584,
+      "step": 755
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003497942386831276,
+      "loss": 0.8333,
+      "step": 760
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034938271604938273,
+      "loss": 0.7459,
+      "step": 765
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003489711934156379,
+      "loss": 0.7499,
+      "step": 770
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.000348559670781893,
+      "loss": 0.7571,
+      "step": 775
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034814814814814816,
+      "loss": 0.7243,
+      "step": 780
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003477366255144033,
+      "loss": 0.7443,
+      "step": 785
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003473251028806585,
+      "loss": 0.7759,
+      "step": 790
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003469135802469136,
+      "loss": 0.7452,
+      "step": 795
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034650205761316874,
+      "loss": 0.7578,
+      "step": 800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003460905349794239,
+      "loss": 0.7634,
+      "step": 805
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.000345679012345679,
+      "loss": 0.7925,
+      "step": 810
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00034526748971193416,
+      "loss": 0.7966,
+      "step": 815
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003448559670781893,
+      "loss": 0.7601,
+      "step": 820
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003444444444444445,
+      "loss": 0.7617,
+      "step": 825
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003440329218106996,
+      "loss": 0.7526,
+      "step": 830
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00034362139917695474,
+      "loss": 0.7297,
+      "step": 835
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003432098765432099,
+      "loss": 0.7549,
+      "step": 840
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.000342798353909465,
+      "loss": 0.7517,
+      "step": 845
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00034238683127572016,
+      "loss": 0.7626,
+      "step": 850
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003419753086419753,
+      "loss": 0.7987,
+      "step": 855
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003415637860082305,
+      "loss": 0.7573,
+      "step": 860
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003411522633744856,
+      "loss": 0.8367,
+      "step": 865
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00034074074074074074,
+      "loss": 0.7258,
+      "step": 870
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003403292181069959,
+      "loss": 0.7632,
+      "step": 875
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00033991769547325106,
+      "loss": 0.7379,
+      "step": 880
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00033950617283950616,
+      "loss": 0.7529,
+      "step": 885
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003390946502057613,
+      "loss": 0.7391,
+      "step": 890
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003386831275720165,
+      "loss": 0.8061,
+      "step": 895
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00033827160493827164,
+      "loss": 0.7288,
+      "step": 900
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003378600823045268,
+      "loss": 0.7493,
+      "step": 905
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00033744855967078195,
+      "loss": 0.7702,
+      "step": 910
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00033703703703703706,
+      "loss": 0.7538,
+      "step": 915
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003366255144032922,
+      "loss": 0.728,
+      "step": 920
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003362139917695474,
+      "loss": 0.7776,
+      "step": 925
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003358024691358025,
+      "loss": 0.7883,
+      "step": 930
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00033539094650205764,
+      "loss": 0.7471,
+      "step": 935
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003349794238683128,
+      "loss": 0.7464,
+      "step": 940
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00033456790123456795,
+      "loss": 0.7885,
+      "step": 945
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00033415637860082306,
+      "loss": 0.7509,
+      "step": 950
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003337448559670782,
+      "loss": 0.7681,
+      "step": 955
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003333333333333334,
+      "loss": 0.8246,
+      "step": 960
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003329218106995885,
+      "loss": 0.7604,
+      "step": 965
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00033251028806584364,
+      "loss": 0.7755,
+      "step": 970
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003320987654320988,
+      "loss": 0.7641,
+      "step": 975
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003316872427983539,
+      "loss": 0.7708,
+      "step": 980
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00033127572016460906,
+      "loss": 0.7936,
+      "step": 985
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003308641975308642,
+      "loss": 0.7641,
+      "step": 990
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003304526748971194,
+      "loss": 0.7479,
+      "step": 995
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003300411522633745,
+      "loss": 0.7547,
+      "step": 1000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00032962962962962964,
+      "loss": 0.762,
+      "step": 1005
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003292181069958848,
+      "loss": 0.7472,
+      "step": 1010
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003288065843621399,
+      "loss": 0.7542,
+      "step": 1015
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00032839506172839506,
+      "loss": 0.7669,
+      "step": 1020
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003279835390946502,
+      "loss": 0.7812,
+      "step": 1025
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003275720164609054,
+      "loss": 0.7273,
+      "step": 1030
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003271604938271605,
+      "loss": 0.7822,
+      "step": 1035
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00032674897119341564,
+      "loss": 0.7337,
+      "step": 1040
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0003263374485596708,
+      "loss": 0.7566,
+      "step": 1045
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0003259259259259259,
+      "loss": 0.7377,
+      "step": 1050
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00032551440329218106,
+      "loss": 0.7856,
+      "step": 1055
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0003251028806584362,
+      "loss": 0.7981,
+      "step": 1060
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0003246913580246914,
+      "loss": 0.7505,
+      "step": 1065
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0003242798353909465,
+      "loss": 0.7549,
+      "step": 1070
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00032386831275720164,
+      "loss": 0.7306,
+      "step": 1075
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0003234567901234568,
+      "loss": 0.7343,
+      "step": 1080
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0003230452674897119,
+      "loss": 0.738,
+      "step": 1085
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00032263374485596707,
+      "loss": 0.7655,
+      "step": 1090
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0003222222222222222,
+      "loss": 0.7706,
+      "step": 1095
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0003218106995884774,
+      "loss": 0.7512,
+      "step": 1100
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00032139917695473254,
+      "loss": 0.7549,
+      "step": 1105
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0003209876543209877,
+      "loss": 0.7839,
+      "step": 1110
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0003205761316872428,
+      "loss": 0.7511,
+      "step": 1115
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00032016460905349796,
+      "loss": 0.769,
+      "step": 1120
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0003197530864197531,
+      "loss": 0.7685,
+      "step": 1125
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0003193415637860083,
+      "loss": 0.7489,
+      "step": 1130
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0003189300411522634,
+      "loss": 0.808,
+      "step": 1135
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00031851851851851854,
+      "loss": 0.7654,
+      "step": 1140
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0003181069958847737,
+      "loss": 0.7845,
+      "step": 1145
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00031769547325102886,
+      "loss": 0.7886,
+      "step": 1150
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00031728395061728396,
+      "loss": 0.7318,
+      "step": 1155
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0003168724279835391,
+      "loss": 0.7231,
+      "step": 1160
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0003164609053497943,
+      "loss": 0.7471,
+      "step": 1165
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0003160493827160494,
+      "loss": 0.7424,
+      "step": 1170
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00031563786008230454,
+      "loss": 0.7249,
+      "step": 1175
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0003152263374485597,
+      "loss": 0.7419,
+      "step": 1180
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00031481481481481486,
+      "loss": 0.7567,
+      "step": 1185
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00031440329218106997,
+      "loss": 0.7707,
+      "step": 1190
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0003139917695473251,
+      "loss": 0.7473,
+      "step": 1195
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0003135802469135803,
+      "loss": 0.7222,
+      "step": 1200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0003131687242798354,
+      "loss": 0.7383,
+      "step": 1205
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00031275720164609055,
+      "loss": 0.7712,
+      "step": 1210
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0003123456790123457,
+      "loss": 0.7602,
+      "step": 1215
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0003119341563786008,
+      "loss": 0.7741,
+      "step": 1220
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00031152263374485597,
+      "loss": 0.7428,
+      "step": 1225
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0003111111111111111,
+      "loss": 0.7605,
+      "step": 1230
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0003106995884773663,
+      "loss": 0.7255,
+      "step": 1235
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0003102880658436214,
+      "loss": 0.7468,
+      "step": 1240
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00030987654320987655,
+      "loss": 0.7961,
+      "step": 1245
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0003094650205761317,
+      "loss": 0.7208,
+      "step": 1250
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0003090534979423868,
+      "loss": 0.7331,
+      "step": 1255
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00030864197530864197,
+      "loss": 0.7371,
+      "step": 1260
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00030823045267489713,
+      "loss": 0.7206,
+      "step": 1265
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0003078189300411523,
+      "loss": 0.7617,
+      "step": 1270
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0003074074074074074,
+      "loss": 0.7652,
+      "step": 1275
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00030699588477366255,
+      "loss": 0.7635,
+      "step": 1280
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0003065843621399177,
+      "loss": 0.7445,
+      "step": 1285
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0003061728395061728,
+      "loss": 0.7717,
+      "step": 1290
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00030576131687242797,
+      "loss": 0.7535,
+      "step": 1295
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00030534979423868313,
+      "loss": 0.7677,
+      "step": 1300
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0003049382716049383,
+      "loss": 0.8131,
+      "step": 1305
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00030452674897119345,
+      "loss": 0.7754,
+      "step": 1310
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00030411522633744855,
+      "loss": 0.76,
+      "step": 1315
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0003037037037037037,
+      "loss": 0.7558,
+      "step": 1320
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00030329218106995887,
+      "loss": 0.7604,
+      "step": 1325
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.000302880658436214,
+      "loss": 0.7716,
+      "step": 1330
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0003024691358024692,
+      "loss": 0.7669,
+      "step": 1335
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0003020576131687243,
+      "loss": 0.7469,
+      "step": 1340
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00030164609053497945,
+      "loss": 0.7511,
+      "step": 1345
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0003012345679012346,
+      "loss": 0.7483,
+      "step": 1350
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00030082304526748976,
+      "loss": 0.7509,
+      "step": 1355
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00030041152263374487,
+      "loss": 0.7249,
+      "step": 1360
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00030000000000000003,
+      "loss": 0.7289,
+      "step": 1365
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002995884773662552,
+      "loss": 0.7605,
+      "step": 1370
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002991769547325103,
+      "loss": 0.7185,
+      "step": 1375
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00029876543209876545,
+      "loss": 0.6974,
+      "step": 1380
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002983539094650206,
+      "loss": 0.7234,
+      "step": 1385
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00029794238683127577,
+      "loss": 0.7506,
+      "step": 1390
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00029753086419753087,
+      "loss": 0.7524,
+      "step": 1395
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00029711934156378603,
+      "loss": 0.7641,
+      "step": 1400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002967078189300412,
+      "loss": 0.6967,
+      "step": 1405
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002962962962962963,
+      "loss": 0.735,
+      "step": 1410
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00029588477366255145,
+      "loss": 0.7685,
+      "step": 1415
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002954732510288066,
+      "loss": 0.758,
+      "step": 1420
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002950617283950617,
+      "loss": 0.763,
+      "step": 1425
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00029465020576131687,
+      "loss": 0.712,
+      "step": 1430
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00029423868312757203,
+      "loss": 0.7236,
+      "step": 1435
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002938271604938272,
+      "loss": 0.7732,
+      "step": 1440
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002934156378600823,
+      "loss": 0.7689,
+      "step": 1445
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00029300411522633745,
+      "loss": 0.7352,
+      "step": 1450
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002925925925925926,
+      "loss": 0.7711,
+      "step": 1455
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002921810699588477,
+      "loss": 0.7549,
+      "step": 1460
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002917695473251029,
+      "loss": 0.7804,
+      "step": 1465
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00029135802469135803,
+      "loss": 0.7478,
+      "step": 1470
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002909465020576132,
+      "loss": 0.7801,
+      "step": 1475
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002905349794238683,
+      "loss": 0.7544,
+      "step": 1480
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00029012345679012345,
+      "loss": 0.7546,
+      "step": 1485
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002897119341563786,
+      "loss": 0.7648,
+      "step": 1490
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002893004115226337,
+      "loss": 0.7468,
+      "step": 1495
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002888888888888889,
+      "loss": 0.7434,
+      "step": 1500
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00028847736625514403,
+      "loss": 0.7348,
+      "step": 1505
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002880658436213992,
+      "loss": 0.7515,
+      "step": 1510
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00028765432098765435,
+      "loss": 0.7345,
+      "step": 1515
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00028724279835390946,
+      "loss": 0.7701,
+      "step": 1520
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0002868312757201646,
+      "loss": 0.7472,
+      "step": 1525
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00028641975308641977,
+      "loss": 0.7075,
+      "step": 1530
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00028600823045267493,
+      "loss": 0.7483,
+      "step": 1535
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0002855967078189301,
+      "loss": 0.712,
+      "step": 1540
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002851851851851852,
+      "loss": 0.7484,
+      "step": 1545
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00028477366255144035,
+      "loss": 0.7471,
+      "step": 1550
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002843621399176955,
+      "loss": 0.7462,
+      "step": 1555
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00028395061728395067,
+      "loss": 0.7371,
+      "step": 1560
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002835390946502058,
+      "loss": 0.7753,
+      "step": 1565
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00028312757201646093,
+      "loss": 0.7374,
+      "step": 1570
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002827160493827161,
+      "loss": 0.7328,
+      "step": 1575
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002823045267489712,
+      "loss": 0.7159,
+      "step": 1580
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00028189300411522635,
+      "loss": 0.7594,
+      "step": 1585
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002814814814814815,
+      "loss": 0.7215,
+      "step": 1590
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00028106995884773667,
+      "loss": 0.7398,
+      "step": 1595
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0002806584362139918,
+      "loss": 0.773,
+      "step": 1600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00028024691358024693,
+      "loss": 0.7584,
+      "step": 1605
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0002798353909465021,
+      "loss": 0.7117,
+      "step": 1610
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0002794238683127572,
+      "loss": 0.7471,
+      "step": 1615
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00027901234567901236,
+      "loss": 0.7443,
+      "step": 1620
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002786008230452675,
+      "loss": 0.7292,
+      "step": 1625
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002781893004115227,
+      "loss": 0.7716,
+      "step": 1630
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002777777777777778,
+      "loss": 0.763,
+      "step": 1635
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00027736625514403294,
+      "loss": 0.7408,
+      "step": 1640
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002769547325102881,
+      "loss": 0.7403,
+      "step": 1645
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002765432098765432,
+      "loss": 0.7508,
+      "step": 1650
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00027613168724279836,
+      "loss": 0.7687,
+      "step": 1655
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002757201646090535,
+      "loss": 0.7305,
+      "step": 1660
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002753086419753086,
+      "loss": 0.7165,
+      "step": 1665
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002748971193415638,
+      "loss": 0.8,
+      "step": 1670
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00027448559670781894,
+      "loss": 0.7357,
+      "step": 1675
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002740740740740741,
+      "loss": 0.7988,
+      "step": 1680
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002736625514403292,
+      "loss": 0.7559,
+      "step": 1685
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00027325102880658436,
+      "loss": 0.712,
+      "step": 1690
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002728395061728395,
+      "loss": 0.7486,
+      "step": 1695
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002724279835390946,
+      "loss": 0.7388,
+      "step": 1700
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002720164609053498,
+      "loss": 0.7441,
+      "step": 1705
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00027160493827160494,
+      "loss": 0.7547,
+      "step": 1710
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002711934156378601,
+      "loss": 0.7229,
+      "step": 1715
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002707818930041152,
+      "loss": 0.7383,
+      "step": 1720
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00027037037037037036,
+      "loss": 0.7606,
+      "step": 1725
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002699588477366255,
+      "loss": 0.7529,
+      "step": 1730
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002695473251028807,
+      "loss": 0.752,
+      "step": 1735
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00026913580246913584,
+      "loss": 0.7342,
+      "step": 1740
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.000268724279835391,
+      "loss": 0.7308,
+      "step": 1745
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002683127572016461,
+      "loss": 0.6915,
+      "step": 1750
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00026790123456790126,
+      "loss": 0.7344,
+      "step": 1755
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002674897119341564,
+      "loss": 0.7426,
+      "step": 1760
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002670781893004116,
+      "loss": 0.7088,
+      "step": 1765
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0002666666666666667,
+      "loss": 0.7241,
+      "step": 1770
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00026625514403292184,
+      "loss": 0.7552,
+      "step": 1775
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.000265843621399177,
+      "loss": 0.7477,
+      "step": 1780
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0002654320987654321,
+      "loss": 0.7115,
+      "step": 1785
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00026502057613168726,
+      "loss": 0.7538,
+      "step": 1790
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0002646090534979424,
+      "loss": 0.7321,
+      "step": 1795
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0002641975308641976,
+      "loss": 0.7212,
+      "step": 1800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0002637860082304527,
+      "loss": 0.747,
+      "step": 1805
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00026337448559670784,
+      "loss": 0.7606,
+      "step": 1810
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.000262962962962963,
+      "loss": 0.7344,
+      "step": 1815
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002625514403292181,
+      "loss": 0.7783,
+      "step": 1820
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00026213991769547326,
+      "loss": 0.7556,
+      "step": 1825
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002617283950617284,
+      "loss": 0.7585,
+      "step": 1830
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002613168724279836,
+      "loss": 0.7359,
+      "step": 1835
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002609053497942387,
+      "loss": 0.7185,
+      "step": 1840
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00026049382716049384,
+      "loss": 0.7573,
+      "step": 1845
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.000260082304526749,
+      "loss": 0.7619,
+      "step": 1850
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0002596707818930041,
+      "loss": 0.7611,
+      "step": 1855
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00025925925925925926,
+      "loss": 0.7647,
+      "step": 1860
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0002588477366255144,
+      "loss": 0.7398,
+      "step": 1865
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0002584362139917695,
+      "loss": 0.7683,
+      "step": 1870
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0002580246913580247,
+      "loss": 0.7817,
+      "step": 1875
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00025761316872427984,
+      "loss": 0.742,
+      "step": 1880
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.000257201646090535,
+      "loss": 0.7509,
+      "step": 1885
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0002567901234567901,
+      "loss": 0.7058,
+      "step": 1890
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00025637860082304526,
+      "loss": 0.7635,
+      "step": 1895
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0002559670781893004,
+      "loss": 0.734,
+      "step": 1900
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00025555555555555553,
+      "loss": 0.7416,
+      "step": 1905
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0002551440329218107,
+      "loss": 0.7071,
+      "step": 1910
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00025473251028806584,
+      "loss": 0.7275,
+      "step": 1915
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.000254320987654321,
+      "loss": 0.7413,
+      "step": 1920
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0002539094650205761,
+      "loss": 0.7703,
+      "step": 1925
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00025349794238683127,
+      "loss": 0.7331,
+      "step": 1930
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0002530864197530864,
+      "loss": 0.7251,
+      "step": 1935
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0002526748971193416,
+      "loss": 0.7341,
+      "step": 1940
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00025226337448559674,
+      "loss": 0.7079,
+      "step": 1945
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00025185185185185185,
+      "loss": 0.7615,
+      "step": 1950
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.000251440329218107,
+      "loss": 0.7144,
+      "step": 1955
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00025102880658436216,
+      "loss": 0.7527,
+      "step": 1960
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0002506172839506173,
+      "loss": 0.7327,
+      "step": 1965
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0002502057613168725,
+      "loss": 0.7173,
+      "step": 1970
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0002497942386831276,
+      "loss": 0.7239,
+      "step": 1975
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00024938271604938274,
+      "loss": 0.7442,
+      "step": 1980
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0002489711934156379,
+      "loss": 0.7396,
+      "step": 1985
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.000248559670781893,
+      "loss": 0.7365,
+      "step": 1990
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00024814814814814816,
+      "loss": 0.7037,
+      "step": 1995
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0002477366255144033,
+      "loss": 0.721,
+      "step": 2000
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0002473251028806585,
+      "loss": 0.7353,
+      "step": 2005
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0002469135802469136,
+      "loss": 0.7648,
+      "step": 2010
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00024650205761316874,
+      "loss": 0.7445,
+      "step": 2015
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0002460905349794239,
+      "loss": 0.7664,
+      "step": 2020
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.000245679012345679,
+      "loss": 0.7202,
+      "step": 2025
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.00024526748971193417,
+      "loss": 0.7147,
+      "step": 2030
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0002448559670781893,
+      "loss": 0.722,
+      "step": 2035
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0002444444444444445,
+      "loss": 0.7539,
+      "step": 2040
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.0002440329218106996,
+      "loss": 0.7681,
+      "step": 2045
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.00024362139917695475,
+      "loss": 0.7304,
+      "step": 2050
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.0002432098765432099,
+      "loss": 0.7432,
+      "step": 2055
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.000242798353909465,
+      "loss": 0.6945,
+      "step": 2060
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.00024238683127572017,
+      "loss": 0.7426,
+      "step": 2065
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00024197530864197533,
+      "loss": 0.7144,
+      "step": 2070
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00024156378600823048,
+      "loss": 0.7761,
+      "step": 2075
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.0002411522633744856,
+      "loss": 0.7402,
+      "step": 2080
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00024074074074074075,
+      "loss": 0.7376,
+      "step": 2085
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.0002403292181069959,
+      "loss": 0.7231,
+      "step": 2090
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00023991769547325104,
+      "loss": 0.7461,
+      "step": 2095
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00023950617283950617,
+      "loss": 0.7215,
+      "step": 2100
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00023909465020576133,
+      "loss": 0.7419,
+      "step": 2105
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00023868312757201646,
+      "loss": 0.7502,
+      "step": 2110
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00023827160493827162,
+      "loss": 0.7294,
+      "step": 2115
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00023786008230452678,
+      "loss": 0.7904,
+      "step": 2120
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00023744855967078194,
+      "loss": 0.7418,
+      "step": 2125
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00023703703703703704,
+      "loss": 0.7371,
+      "step": 2130
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.0002366255144032922,
+      "loss": 0.7384,
+      "step": 2135
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00023621399176954736,
+      "loss": 0.7552,
+      "step": 2140
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00023580246913580246,
+      "loss": 0.7516,
+      "step": 2145
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00023539094650205762,
+      "loss": 0.7303,
+      "step": 2150
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00023497942386831278,
+      "loss": 0.7496,
+      "step": 2155
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00023456790123456794,
+      "loss": 0.7135,
+      "step": 2160
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00023415637860082304,
+      "loss": 0.7283,
+      "step": 2165
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.0002337448559670782,
+      "loss": 0.751,
+      "step": 2170
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00023333333333333336,
+      "loss": 0.7183,
+      "step": 2175
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00023292181069958846,
+      "loss": 0.7242,
+      "step": 2180
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00023251028806584362,
+      "loss": 0.725,
+      "step": 2185
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00023209876543209878,
+      "loss": 0.753,
+      "step": 2190
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0002316872427983539,
+      "loss": 0.7664,
+      "step": 2195
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.00023127572016460904,
+      "loss": 0.7444,
+      "step": 2200
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0002308641975308642,
+      "loss": 0.7226,
+      "step": 2205
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.00023045267489711936,
+      "loss": 0.7653,
+      "step": 2210
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0002300411522633745,
+      "loss": 0.7382,
+      "step": 2215
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00022962962962962965,
+      "loss": 0.7242,
+      "step": 2220
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.0002292181069958848,
+      "loss": 0.7215,
+      "step": 2225
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.0002288065843621399,
+      "loss": 0.724,
+      "step": 2230
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00022839506172839507,
+      "loss": 0.7439,
+      "step": 2235
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00022798353909465023,
+      "loss": 0.699,
+      "step": 2240
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0002275720164609054,
+      "loss": 0.7602,
+      "step": 2245
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0002271604938271605,
+      "loss": 0.7103,
+      "step": 2250
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.00022674897119341565,
+      "loss": 0.7015,
+      "step": 2255
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0002263374485596708,
+      "loss": 0.7404,
+      "step": 2260
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.00022592592592592591,
+      "loss": 0.7127,
+      "step": 2265
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.00022551440329218107,
+      "loss": 0.7299,
+      "step": 2270
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.00022510288065843623,
+      "loss": 0.7174,
+      "step": 2275
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0002246913580246914,
+      "loss": 0.7454,
+      "step": 2280
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0002242798353909465,
+      "loss": 0.7159,
+      "step": 2285
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.00022386831275720165,
+      "loss": 0.7263,
+      "step": 2290
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.0002234567901234568,
+      "loss": 0.7148,
+      "step": 2295
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00022304526748971192,
+      "loss": 0.7447,
+      "step": 2300
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00022263374485596707,
+      "loss": 0.7326,
+      "step": 2305
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00022222222222222223,
+      "loss": 0.7428,
+      "step": 2310
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00022181069958847736,
+      "loss": 0.7587,
+      "step": 2315
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.00022139917695473252,
+      "loss": 0.759,
+      "step": 2320
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.00022098765432098768,
+      "loss": 0.7716,
+      "step": 2325
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0002205761316872428,
+      "loss": 0.74,
+      "step": 2330
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.00022016460905349794,
+      "loss": 0.7255,
+      "step": 2335
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0002197530864197531,
+      "loss": 0.7547,
+      "step": 2340
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00021934156378600826,
+      "loss": 0.7359,
+      "step": 2345
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00021893004115226337,
+      "loss": 0.6987,
+      "step": 2350
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00021851851851851852,
+      "loss": 0.7233,
+      "step": 2355
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00021810699588477368,
+      "loss": 0.6945,
+      "step": 2360
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00021769547325102884,
+      "loss": 0.7566,
+      "step": 2365
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00021728395061728395,
+      "loss": 0.728,
+      "step": 2370
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.0002168724279835391,
+      "loss": 0.7167,
+      "step": 2375
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00021646090534979426,
+      "loss": 0.7576,
+      "step": 2380
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00021604938271604937,
+      "loss": 0.7246,
+      "step": 2385
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00021563786008230453,
+      "loss": 0.755,
+      "step": 2390
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00021522633744855968,
+      "loss": 0.6983,
+      "step": 2395
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00021481481481481484,
+      "loss": 0.7369,
+      "step": 2400
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00021440329218106995,
+      "loss": 0.7074,
+      "step": 2405
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0002139917695473251,
+      "loss": 0.7317,
+      "step": 2410
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00021358024691358026,
+      "loss": 0.7086,
+      "step": 2415
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.0002131687242798354,
+      "loss": 0.7356,
+      "step": 2420
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00021275720164609055,
+      "loss": 0.7652,
+      "step": 2425
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00021234567901234569,
+      "loss": 0.7488,
+      "step": 2430
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00021193415637860082,
+      "loss": 0.746,
+      "step": 2435
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00021152263374485598,
+      "loss": 0.7454,
+      "step": 2440
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00021111111111111113,
+      "loss": 0.72,
+      "step": 2445
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0002106995884773663,
+      "loss": 0.7299,
+      "step": 2450
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0002102880658436214,
+      "loss": 0.7436,
+      "step": 2455
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00020987654320987656,
+      "loss": 0.7068,
+      "step": 2460
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00020946502057613171,
+      "loss": 0.711,
+      "step": 2465
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.00020905349794238682,
+      "loss": 0.7473,
+      "step": 2470
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.00020864197530864198,
+      "loss": 0.7016,
+      "step": 2475
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.00020823045267489714,
+      "loss": 0.7049,
+      "step": 2480
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0002078189300411523,
+      "loss": 0.7268,
+      "step": 2485
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0002074074074074074,
+      "loss": 0.7239,
+      "step": 2490
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.00020699588477366256,
+      "loss": 0.698,
+      "step": 2495
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.00020658436213991772,
+      "loss": 0.7573,
+      "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_4e4_s11_1/checkpoint-2500/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43458f4cf6aeb211c730b848f93ab15f483827c314fb4dc1e79dd1f658b7dbc8
+size 5944

save_lora_instruct_r8_4e4_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_4e4_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_4e4_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_4e4_s11_1/checkpoint-3000/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:594b67f8ef6310c9ff1be2eeb1ef6e747610cd7066b4afb6e1565107960ae791
+size 564342240

save_lora_instruct_r8_4e4_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 sha256:9040a90f5d8818966640769ae8b1972e8162567456b23ae9d6b84ef4032e980b
+size 239914224

save_lora_instruct_r8_4e4_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:fea4b1fe3e6d7ecadb081b903d9a5b6f21d8dd89b942f105211beee91c283886
+size 40303404

save_lora_instruct_r8_4e4_s11_1/checkpoint-3000/latest

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

save_lora_instruct_r8_4e4_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_4e4_s11_1/checkpoint-3000/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:083abc970a85acd0fe0036aee21d5b5cee2ceb437d75185f43bb0e09b5a91147
+size 1064

save_lora_instruct_r8_4e4_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": 1.3333333333333333e-05,
+      "loss": 1.3798,
+      "step": 5
+    },
+    {
+      "epoch": 0.0,
+      "learning_rate": 2.6666666666666667e-05,
+      "loss": 1.3917,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 4e-05,
+      "loss": 1.2862,
+      "step": 15
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 5.333333333333333e-05,
+      "loss": 1.2046,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 6.666666666666667e-05,
+      "loss": 1.0634,
+      "step": 25
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 8e-05,
+      "loss": 1.0273,
+      "step": 30
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 9.333333333333334e-05,
+      "loss": 1.0472,
+      "step": 35
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00010666666666666667,
+      "loss": 1.0301,
+      "step": 40
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00012,
+      "loss": 0.9335,
+      "step": 45
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00013333333333333334,
+      "loss": 0.921,
+      "step": 50
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00014666666666666666,
+      "loss": 0.9506,
+      "step": 55
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 0.00016,
+      "loss": 0.8876,
+      "step": 60
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00017333333333333334,
+      "loss": 0.9335,
+      "step": 65
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0001866666666666667,
+      "loss": 0.8554,
+      "step": 70
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.0002,
+      "loss": 0.8537,
+      "step": 75
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00021333333333333333,
+      "loss": 0.9033,
+      "step": 80
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 0.00022666666666666668,
+      "loss": 0.8203,
+      "step": 85
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00024,
+      "loss": 0.8611,
+      "step": 90
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00025333333333333333,
+      "loss": 0.8303,
+      "step": 95
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0002666666666666667,
+      "loss": 0.8528,
+      "step": 100
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.00028,
+      "loss": 0.8537,
+      "step": 105
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 0.0002933333333333333,
+      "loss": 0.8286,
+      "step": 110
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0003066666666666667,
+      "loss": 0.8449,
+      "step": 115
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00032,
+      "loss": 0.8377,
+      "step": 120
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.0003333333333333334,
+      "loss": 0.8419,
+      "step": 125
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00034666666666666667,
+      "loss": 0.8265,
+      "step": 130
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 0.00036,
+      "loss": 0.8037,
+      "step": 135
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0003733333333333334,
+      "loss": 0.7924,
+      "step": 140
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.00038666666666666667,
+      "loss": 0.8245,
+      "step": 145
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0004,
+      "loss": 0.8388,
+      "step": 150
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0003995884773662552,
+      "loss": 0.8066,
+      "step": 155
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 0.0003991769547325103,
+      "loss": 0.8107,
+      "step": 160
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00039876543209876544,
+      "loss": 0.8005,
+      "step": 165
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.0003983539094650206,
+      "loss": 0.7727,
+      "step": 170
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00039794238683127576,
+      "loss": 0.8276,
+      "step": 175
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.00039753086419753086,
+      "loss": 0.8464,
+      "step": 180
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 0.000397119341563786,
+      "loss": 0.8348,
+      "step": 185
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0003967078189300412,
+      "loss": 0.8228,
+      "step": 190
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0003962962962962963,
+      "loss": 0.8254,
+      "step": 195
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.00039588477366255144,
+      "loss": 0.817,
+      "step": 200
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0003954732510288066,
+      "loss": 0.8365,
+      "step": 205
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 0.0003950617283950617,
+      "loss": 0.793,
+      "step": 210
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00039465020576131686,
+      "loss": 0.8014,
+      "step": 215
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.000394238683127572,
+      "loss": 0.8049,
+      "step": 220
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0003938271604938272,
+      "loss": 0.7841,
+      "step": 225
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.0003934156378600823,
+      "loss": 0.7977,
+      "step": 230
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 0.00039300411522633744,
+      "loss": 0.8299,
+      "step": 235
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0003925925925925926,
+      "loss": 0.7729,
+      "step": 240
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.00039218106995884776,
+      "loss": 0.8082,
+      "step": 245
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0003917695473251029,
+      "loss": 0.8173,
+      "step": 250
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.000391358024691358,
+      "loss": 0.8379,
+      "step": 255
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 0.0003909465020576132,
+      "loss": 0.7962,
+      "step": 260
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00039053497942386834,
+      "loss": 0.8067,
+      "step": 265
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0003901234567901235,
+      "loss": 0.7915,
+      "step": 270
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00038971193415637866,
+      "loss": 0.7905,
+      "step": 275
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.00038930041152263376,
+      "loss": 0.7862,
+      "step": 280
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 0.0003888888888888889,
+      "loss": 0.7709,
+      "step": 285
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0003884773662551441,
+      "loss": 0.7773,
+      "step": 290
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00038806584362139924,
+      "loss": 0.8068,
+      "step": 295
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00038765432098765434,
+      "loss": 0.7962,
+      "step": 300
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.0003872427983539095,
+      "loss": 0.7894,
+      "step": 305
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 0.00038683127572016466,
+      "loss": 0.7705,
+      "step": 310
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00038641975308641976,
+      "loss": 0.7882,
+      "step": 315
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0003860082304526749,
+      "loss": 0.7865,
+      "step": 320
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0003855967078189301,
+      "loss": 0.7939,
+      "step": 325
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.0003851851851851852,
+      "loss": 0.7724,
+      "step": 330
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 0.00038477366255144034,
+      "loss": 0.7771,
+      "step": 335
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0003843621399176955,
+      "loss": 0.7941,
+      "step": 340
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00038395061728395066,
+      "loss": 0.7822,
+      "step": 345
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.00038353909465020577,
+      "loss": 0.8143,
+      "step": 350
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0003831275720164609,
+      "loss": 0.795,
+      "step": 355
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 0.0003827160493827161,
+      "loss": 0.8049,
+      "step": 360
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0003823045267489712,
+      "loss": 0.8309,
+      "step": 365
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00038189300411522635,
+      "loss": 0.8314,
+      "step": 370
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.0003814814814814815,
+      "loss": 0.801,
+      "step": 375
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00038106995884773666,
+      "loss": 0.8084,
+      "step": 380
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 0.00038065843621399177,
+      "loss": 0.8129,
+      "step": 385
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0003802469135802469,
+      "loss": 0.7859,
+      "step": 390
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0003798353909465021,
+      "loss": 0.7874,
+      "step": 395
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0003794238683127572,
+      "loss": 0.7548,
+      "step": 400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.00037901234567901235,
+      "loss": 0.8092,
+      "step": 405
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 0.0003786008230452675,
+      "loss": 0.8041,
+      "step": 410
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00037818930041152266,
+      "loss": 0.777,
+      "step": 415
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00037777777777777777,
+      "loss": 0.7699,
+      "step": 420
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.00037736625514403293,
+      "loss": 0.781,
+      "step": 425
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0003769547325102881,
+      "loss": 0.7504,
+      "step": 430
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 0.0003765432098765432,
+      "loss": 0.7974,
+      "step": 435
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00037613168724279835,
+      "loss": 0.7663,
+      "step": 440
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.0003757201646090535,
+      "loss": 0.8013,
+      "step": 445
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00037530864197530867,
+      "loss": 0.7453,
+      "step": 450
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00037489711934156377,
+      "loss": 0.7425,
+      "step": 455
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 0.00037448559670781893,
+      "loss": 0.7582,
+      "step": 460
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0003740740740740741,
+      "loss": 0.7801,
+      "step": 465
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00037366255144032925,
+      "loss": 0.7809,
+      "step": 470
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.0003732510288065844,
+      "loss": 0.7749,
+      "step": 475
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00037283950617283956,
+      "loss": 0.7807,
+      "step": 480
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 0.00037242798353909467,
+      "loss": 0.7715,
+      "step": 485
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0003720164609053498,
+      "loss": 0.7356,
+      "step": 490
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.000371604938271605,
+      "loss": 0.7674,
+      "step": 495
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00037119341563786014,
+      "loss": 0.7542,
+      "step": 500
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.00037078189300411525,
+      "loss": 0.7583,
+      "step": 505
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 0.0003703703703703704,
+      "loss": 0.7767,
+      "step": 510
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00036995884773662556,
+      "loss": 0.7853,
+      "step": 515
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00036954732510288067,
+      "loss": 0.7733,
+      "step": 520
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00036913580246913583,
+      "loss": 0.7758,
+      "step": 525
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.000368724279835391,
+      "loss": 0.7698,
+      "step": 530
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 0.00036831275720164614,
+      "loss": 0.7805,
+      "step": 535
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00036790123456790125,
+      "loss": 0.7573,
+      "step": 540
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.0003674897119341564,
+      "loss": 0.7709,
+      "step": 545
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00036707818930041157,
+      "loss": 0.7994,
+      "step": 550
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00036666666666666667,
+      "loss": 0.7363,
+      "step": 555
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 0.00036625514403292183,
+      "loss": 0.7478,
+      "step": 560
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.000365843621399177,
+      "loss": 0.8188,
+      "step": 565
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003654320987654321,
+      "loss": 0.7524,
+      "step": 570
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00036502057613168725,
+      "loss": 0.7319,
+      "step": 575
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.0003646090534979424,
+      "loss": 0.7915,
+      "step": 580
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 0.00036419753086419757,
+      "loss": 0.7722,
+      "step": 585
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00036378600823045267,
+      "loss": 0.7842,
+      "step": 590
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00036337448559670783,
+      "loss": 0.7591,
+      "step": 595
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.000362962962962963,
+      "loss": 0.7653,
+      "step": 600
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.0003625514403292181,
+      "loss": 0.7838,
+      "step": 605
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 0.00036213991769547325,
+      "loss": 0.7937,
+      "step": 610
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003617283950617284,
+      "loss": 0.77,
+      "step": 615
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00036131687242798357,
+      "loss": 0.7899,
+      "step": 620
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.0003609053497942387,
+      "loss": 0.7815,
+      "step": 625
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.00036049382716049383,
+      "loss": 0.7648,
+      "step": 630
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 0.000360082304526749,
+      "loss": 0.7716,
+      "step": 635
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003596707818930041,
+      "loss": 0.7708,
+      "step": 640
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.00035925925925925925,
+      "loss": 0.7306,
+      "step": 645
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003588477366255144,
+      "loss": 0.8007,
+      "step": 650
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003584362139917695,
+      "loss": 0.7551,
+      "step": 655
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 0.0003580246913580247,
+      "loss": 0.754,
+      "step": 660
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00035761316872427983,
+      "loss": 0.7599,
+      "step": 665
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.000357201646090535,
+      "loss": 0.7637,
+      "step": 670
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.00035679012345679015,
+      "loss": 0.7766,
+      "step": 675
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003563786008230453,
+      "loss": 0.7727,
+      "step": 680
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 0.0003559670781893004,
+      "loss": 0.7675,
+      "step": 685
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00035555555555555557,
+      "loss": 0.7707,
+      "step": 690
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00035514403292181073,
+      "loss": 0.7221,
+      "step": 695
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.0003547325102880659,
+      "loss": 0.7335,
+      "step": 700
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00035432098765432105,
+      "loss": 0.781,
+      "step": 705
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 0.00035390946502057615,
+      "loss": 0.7845,
+      "step": 710
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003534979423868313,
+      "loss": 0.7467,
+      "step": 715
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035308641975308647,
+      "loss": 0.7735,
+      "step": 720
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003526748971193416,
+      "loss": 0.7368,
+      "step": 725
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.00035226337448559673,
+      "loss": 0.7678,
+      "step": 730
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 0.0003518518518518519,
+      "loss": 0.7756,
+      "step": 735
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035144032921810705,
+      "loss": 0.7635,
+      "step": 740
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035102880658436215,
+      "loss": 0.7757,
+      "step": 745
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003506172839506173,
+      "loss": 0.7442,
+      "step": 750
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.00035020576131687247,
+      "loss": 0.7584,
+      "step": 755
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 0.0003497942386831276,
+      "loss": 0.8333,
+      "step": 760
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034938271604938273,
+      "loss": 0.7459,
+      "step": 765
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003489711934156379,
+      "loss": 0.7499,
+      "step": 770
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.000348559670781893,
+      "loss": 0.7571,
+      "step": 775
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.00034814814814814816,
+      "loss": 0.7243,
+      "step": 780
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 0.0003477366255144033,
+      "loss": 0.7443,
+      "step": 785
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003473251028806585,
+      "loss": 0.7759,
+      "step": 790
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003469135802469136,
+      "loss": 0.7452,
+      "step": 795
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.00034650205761316874,
+      "loss": 0.7578,
+      "step": 800
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.0003460905349794239,
+      "loss": 0.7634,
+      "step": 805
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 0.000345679012345679,
+      "loss": 0.7925,
+      "step": 810
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00034526748971193416,
+      "loss": 0.7966,
+      "step": 815
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003448559670781893,
+      "loss": 0.7601,
+      "step": 820
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003444444444444445,
+      "loss": 0.7617,
+      "step": 825
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.0003440329218106996,
+      "loss": 0.7526,
+      "step": 830
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 0.00034362139917695474,
+      "loss": 0.7297,
+      "step": 835
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003432098765432099,
+      "loss": 0.7549,
+      "step": 840
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.000342798353909465,
+      "loss": 0.7517,
+      "step": 845
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.00034238683127572016,
+      "loss": 0.7626,
+      "step": 850
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003419753086419753,
+      "loss": 0.7987,
+      "step": 855
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 0.0003415637860082305,
+      "loss": 0.7573,
+      "step": 860
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003411522633744856,
+      "loss": 0.8367,
+      "step": 865
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00034074074074074074,
+      "loss": 0.7258,
+      "step": 870
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.0003403292181069959,
+      "loss": 0.7632,
+      "step": 875
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00033991769547325106,
+      "loss": 0.7379,
+      "step": 880
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 0.00033950617283950616,
+      "loss": 0.7529,
+      "step": 885
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003390946502057613,
+      "loss": 0.7391,
+      "step": 890
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003386831275720165,
+      "loss": 0.8061,
+      "step": 895
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00033827160493827164,
+      "loss": 0.7288,
+      "step": 900
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.0003378600823045268,
+      "loss": 0.7493,
+      "step": 905
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 0.00033744855967078195,
+      "loss": 0.7702,
+      "step": 910
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00033703703703703706,
+      "loss": 0.7538,
+      "step": 915
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003366255144032922,
+      "loss": 0.728,
+      "step": 920
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003362139917695474,
+      "loss": 0.7776,
+      "step": 925
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.0003358024691358025,
+      "loss": 0.7883,
+      "step": 930
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 0.00033539094650205764,
+      "loss": 0.7471,
+      "step": 935
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003349794238683128,
+      "loss": 0.7464,
+      "step": 940
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00033456790123456795,
+      "loss": 0.7885,
+      "step": 945
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.00033415637860082306,
+      "loss": 0.7509,
+      "step": 950
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003337448559670782,
+      "loss": 0.7681,
+      "step": 955
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 0.0003333333333333334,
+      "loss": 0.8246,
+      "step": 960
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003329218106995885,
+      "loss": 0.7604,
+      "step": 965
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00033251028806584364,
+      "loss": 0.7755,
+      "step": 970
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003320987654320988,
+      "loss": 0.7641,
+      "step": 975
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.0003316872427983539,
+      "loss": 0.7708,
+      "step": 980
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 0.00033127572016460906,
+      "loss": 0.7936,
+      "step": 985
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003308641975308642,
+      "loss": 0.7641,
+      "step": 990
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003304526748971194,
+      "loss": 0.7479,
+      "step": 995
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003300411522633745,
+      "loss": 0.7547,
+      "step": 1000
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.00032962962962962964,
+      "loss": 0.762,
+      "step": 1005
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 0.0003292181069958848,
+      "loss": 0.7472,
+      "step": 1010
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003288065843621399,
+      "loss": 0.7542,
+      "step": 1015
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.00032839506172839506,
+      "loss": 0.7669,
+      "step": 1020
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003279835390946502,
+      "loss": 0.7812,
+      "step": 1025
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003275720164609054,
+      "loss": 0.7273,
+      "step": 1030
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 0.0003271604938271605,
+      "loss": 0.7822,
+      "step": 1035
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00032674897119341564,
+      "loss": 0.7337,
+      "step": 1040
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0003263374485596708,
+      "loss": 0.7566,
+      "step": 1045
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0003259259259259259,
+      "loss": 0.7377,
+      "step": 1050
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.00032551440329218106,
+      "loss": 0.7856,
+      "step": 1055
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 0.0003251028806584362,
+      "loss": 0.7981,
+      "step": 1060
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0003246913580246914,
+      "loss": 0.7505,
+      "step": 1065
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0003242798353909465,
+      "loss": 0.7549,
+      "step": 1070
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.00032386831275720164,
+      "loss": 0.7306,
+      "step": 1075
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0003234567901234568,
+      "loss": 0.7343,
+      "step": 1080
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 0.0003230452674897119,
+      "loss": 0.738,
+      "step": 1085
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00032263374485596707,
+      "loss": 0.7655,
+      "step": 1090
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0003222222222222222,
+      "loss": 0.7706,
+      "step": 1095
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0003218106995884774,
+      "loss": 0.7512,
+      "step": 1100
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.00032139917695473254,
+      "loss": 0.7549,
+      "step": 1105
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 0.0003209876543209877,
+      "loss": 0.7839,
+      "step": 1110
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0003205761316872428,
+      "loss": 0.7511,
+      "step": 1115
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.00032016460905349796,
+      "loss": 0.769,
+      "step": 1120
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0003197530864197531,
+      "loss": 0.7685,
+      "step": 1125
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0003193415637860083,
+      "loss": 0.7489,
+      "step": 1130
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 0.0003189300411522634,
+      "loss": 0.808,
+      "step": 1135
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00031851851851851854,
+      "loss": 0.7654,
+      "step": 1140
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0003181069958847737,
+      "loss": 0.7845,
+      "step": 1145
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00031769547325102886,
+      "loss": 0.7886,
+      "step": 1150
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.00031728395061728396,
+      "loss": 0.7318,
+      "step": 1155
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 0.0003168724279835391,
+      "loss": 0.7231,
+      "step": 1160
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0003164609053497943,
+      "loss": 0.7471,
+      "step": 1165
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0003160493827160494,
+      "loss": 0.7424,
+      "step": 1170
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00031563786008230454,
+      "loss": 0.7249,
+      "step": 1175
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.0003152263374485597,
+      "loss": 0.7419,
+      "step": 1180
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 0.00031481481481481486,
+      "loss": 0.7567,
+      "step": 1185
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00031440329218106997,
+      "loss": 0.7707,
+      "step": 1190
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0003139917695473251,
+      "loss": 0.7473,
+      "step": 1195
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0003135802469135803,
+      "loss": 0.7222,
+      "step": 1200
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0003131687242798354,
+      "loss": 0.7383,
+      "step": 1205
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.00031275720164609055,
+      "loss": 0.7712,
+      "step": 1210
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 0.0003123456790123457,
+      "loss": 0.7602,
+      "step": 1215
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0003119341563786008,
+      "loss": 0.7741,
+      "step": 1220
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.00031152263374485597,
+      "loss": 0.7428,
+      "step": 1225
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0003111111111111111,
+      "loss": 0.7605,
+      "step": 1230
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0003106995884773663,
+      "loss": 0.7255,
+      "step": 1235
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 0.0003102880658436214,
+      "loss": 0.7468,
+      "step": 1240
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00030987654320987655,
+      "loss": 0.7961,
+      "step": 1245
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0003094650205761317,
+      "loss": 0.7208,
+      "step": 1250
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.0003090534979423868,
+      "loss": 0.7331,
+      "step": 1255
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00030864197530864197,
+      "loss": 0.7371,
+      "step": 1260
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 0.00030823045267489713,
+      "loss": 0.7206,
+      "step": 1265
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0003078189300411523,
+      "loss": 0.7617,
+      "step": 1270
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0003074074074074074,
+      "loss": 0.7652,
+      "step": 1275
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.00030699588477366255,
+      "loss": 0.7635,
+      "step": 1280
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0003065843621399177,
+      "loss": 0.7445,
+      "step": 1285
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 0.0003061728395061728,
+      "loss": 0.7717,
+      "step": 1290
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00030576131687242797,
+      "loss": 0.7535,
+      "step": 1295
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00030534979423868313,
+      "loss": 0.7677,
+      "step": 1300
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.0003049382716049383,
+      "loss": 0.8131,
+      "step": 1305
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00030452674897119345,
+      "loss": 0.7754,
+      "step": 1310
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 0.00030411522633744855,
+      "loss": 0.76,
+      "step": 1315
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0003037037037037037,
+      "loss": 0.7558,
+      "step": 1320
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.00030329218106995887,
+      "loss": 0.7604,
+      "step": 1325
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.000302880658436214,
+      "loss": 0.7716,
+      "step": 1330
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0003024691358024692,
+      "loss": 0.7669,
+      "step": 1335
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 0.0003020576131687243,
+      "loss": 0.7469,
+      "step": 1340
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00030164609053497945,
+      "loss": 0.7511,
+      "step": 1345
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.0003012345679012346,
+      "loss": 0.7483,
+      "step": 1350
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00030082304526748976,
+      "loss": 0.7509,
+      "step": 1355
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00030041152263374487,
+      "loss": 0.7249,
+      "step": 1360
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 0.00030000000000000003,
+      "loss": 0.7289,
+      "step": 1365
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002995884773662552,
+      "loss": 0.7605,
+      "step": 1370
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002991769547325103,
+      "loss": 0.7185,
+      "step": 1375
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00029876543209876545,
+      "loss": 0.6974,
+      "step": 1380
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.0002983539094650206,
+      "loss": 0.7234,
+      "step": 1385
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 0.00029794238683127577,
+      "loss": 0.7506,
+      "step": 1390
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00029753086419753087,
+      "loss": 0.7524,
+      "step": 1395
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00029711934156378603,
+      "loss": 0.7641,
+      "step": 1400
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002967078189300412,
+      "loss": 0.6967,
+      "step": 1405
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.0002962962962962963,
+      "loss": 0.735,
+      "step": 1410
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 0.00029588477366255145,
+      "loss": 0.7685,
+      "step": 1415
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002954732510288066,
+      "loss": 0.758,
+      "step": 1420
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002950617283950617,
+      "loss": 0.763,
+      "step": 1425
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00029465020576131687,
+      "loss": 0.712,
+      "step": 1430
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.00029423868312757203,
+      "loss": 0.7236,
+      "step": 1435
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 0.0002938271604938272,
+      "loss": 0.7732,
+      "step": 1440
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002934156378600823,
+      "loss": 0.7689,
+      "step": 1445
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.00029300411522633745,
+      "loss": 0.7352,
+      "step": 1450
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002925925925925926,
+      "loss": 0.7711,
+      "step": 1455
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002921810699588477,
+      "loss": 0.7549,
+      "step": 1460
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 0.0002917695473251029,
+      "loss": 0.7804,
+      "step": 1465
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00029135802469135803,
+      "loss": 0.7478,
+      "step": 1470
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002909465020576132,
+      "loss": 0.7801,
+      "step": 1475
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002905349794238683,
+      "loss": 0.7544,
+      "step": 1480
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.00029012345679012345,
+      "loss": 0.7546,
+      "step": 1485
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 0.0002897119341563786,
+      "loss": 0.7648,
+      "step": 1490
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002893004115226337,
+      "loss": 0.7468,
+      "step": 1495
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002888888888888889,
+      "loss": 0.7434,
+      "step": 1500
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00028847736625514403,
+      "loss": 0.7348,
+      "step": 1505
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.0002880658436213992,
+      "loss": 0.7515,
+      "step": 1510
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 0.00028765432098765435,
+      "loss": 0.7345,
+      "step": 1515
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00028724279835390946,
+      "loss": 0.7701,
+      "step": 1520
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0002868312757201646,
+      "loss": 0.7472,
+      "step": 1525
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00028641975308641977,
+      "loss": 0.7075,
+      "step": 1530
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.00028600823045267493,
+      "loss": 0.7483,
+      "step": 1535
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 0.0002855967078189301,
+      "loss": 0.712,
+      "step": 1540
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002851851851851852,
+      "loss": 0.7484,
+      "step": 1545
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00028477366255144035,
+      "loss": 0.7471,
+      "step": 1550
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002843621399176955,
+      "loss": 0.7462,
+      "step": 1555
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.00028395061728395067,
+      "loss": 0.7371,
+      "step": 1560
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 0.0002835390946502058,
+      "loss": 0.7753,
+      "step": 1565
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00028312757201646093,
+      "loss": 0.7374,
+      "step": 1570
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002827160493827161,
+      "loss": 0.7328,
+      "step": 1575
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002823045267489712,
+      "loss": 0.7159,
+      "step": 1580
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.00028189300411522635,
+      "loss": 0.7594,
+      "step": 1585
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 0.0002814814814814815,
+      "loss": 0.7215,
+      "step": 1590
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00028106995884773667,
+      "loss": 0.7398,
+      "step": 1595
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0002806584362139918,
+      "loss": 0.773,
+      "step": 1600
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.00028024691358024693,
+      "loss": 0.7584,
+      "step": 1605
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0002798353909465021,
+      "loss": 0.7117,
+      "step": 1610
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 0.0002794238683127572,
+      "loss": 0.7471,
+      "step": 1615
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00027901234567901236,
+      "loss": 0.7443,
+      "step": 1620
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002786008230452675,
+      "loss": 0.7292,
+      "step": 1625
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002781893004115227,
+      "loss": 0.7716,
+      "step": 1630
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.0002777777777777778,
+      "loss": 0.763,
+      "step": 1635
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 0.00027736625514403294,
+      "loss": 0.7408,
+      "step": 1640
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002769547325102881,
+      "loss": 0.7403,
+      "step": 1645
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002765432098765432,
+      "loss": 0.7508,
+      "step": 1650
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.00027613168724279836,
+      "loss": 0.7687,
+      "step": 1655
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002757201646090535,
+      "loss": 0.7305,
+      "step": 1660
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 0.0002753086419753086,
+      "loss": 0.7165,
+      "step": 1665
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002748971193415638,
+      "loss": 0.8,
+      "step": 1670
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00027448559670781894,
+      "loss": 0.7357,
+      "step": 1675
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002740740740740741,
+      "loss": 0.7988,
+      "step": 1680
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.0002736625514403292,
+      "loss": 0.7559,
+      "step": 1685
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 0.00027325102880658436,
+      "loss": 0.712,
+      "step": 1690
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002728395061728395,
+      "loss": 0.7486,
+      "step": 1695
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002724279835390946,
+      "loss": 0.7388,
+      "step": 1700
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002720164609053498,
+      "loss": 0.7441,
+      "step": 1705
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.00027160493827160494,
+      "loss": 0.7547,
+      "step": 1710
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 0.0002711934156378601,
+      "loss": 0.7229,
+      "step": 1715
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002707818930041152,
+      "loss": 0.7383,
+      "step": 1720
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00027037037037037036,
+      "loss": 0.7606,
+      "step": 1725
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002699588477366255,
+      "loss": 0.7529,
+      "step": 1730
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.0002695473251028807,
+      "loss": 0.752,
+      "step": 1735
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 0.00026913580246913584,
+      "loss": 0.7342,
+      "step": 1740
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.000268724279835391,
+      "loss": 0.7308,
+      "step": 1745
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002683127572016461,
+      "loss": 0.6915,
+      "step": 1750
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.00026790123456790126,
+      "loss": 0.7344,
+      "step": 1755
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002674897119341564,
+      "loss": 0.7426,
+      "step": 1760
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 0.0002670781893004116,
+      "loss": 0.7088,
+      "step": 1765
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0002666666666666667,
+      "loss": 0.7241,
+      "step": 1770
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00026625514403292184,
+      "loss": 0.7552,
+      "step": 1775
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.000265843621399177,
+      "loss": 0.7477,
+      "step": 1780
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.0002654320987654321,
+      "loss": 0.7115,
+      "step": 1785
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 0.00026502057613168726,
+      "loss": 0.7538,
+      "step": 1790
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0002646090534979424,
+      "loss": 0.7321,
+      "step": 1795
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0002641975308641976,
+      "loss": 0.7212,
+      "step": 1800
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.0002637860082304527,
+      "loss": 0.747,
+      "step": 1805
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.00026337448559670784,
+      "loss": 0.7606,
+      "step": 1810
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 0.000262962962962963,
+      "loss": 0.7344,
+      "step": 1815
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002625514403292181,
+      "loss": 0.7783,
+      "step": 1820
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.00026213991769547326,
+      "loss": 0.7556,
+      "step": 1825
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002617283950617284,
+      "loss": 0.7585,
+      "step": 1830
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002613168724279836,
+      "loss": 0.7359,
+      "step": 1835
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 0.0002609053497942387,
+      "loss": 0.7185,
+      "step": 1840
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00026049382716049384,
+      "loss": 0.7573,
+      "step": 1845
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.000260082304526749,
+      "loss": 0.7619,
+      "step": 1850
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0002596707818930041,
+      "loss": 0.7611,
+      "step": 1855
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.00025925925925925926,
+      "loss": 0.7647,
+      "step": 1860
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 0.0002588477366255144,
+      "loss": 0.7398,
+      "step": 1865
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0002584362139917695,
+      "loss": 0.7683,
+      "step": 1870
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0002580246913580247,
+      "loss": 0.7817,
+      "step": 1875
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.00025761316872427984,
+      "loss": 0.742,
+      "step": 1880
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.000257201646090535,
+      "loss": 0.7509,
+      "step": 1885
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 0.0002567901234567901,
+      "loss": 0.7058,
+      "step": 1890
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00025637860082304526,
+      "loss": 0.7635,
+      "step": 1895
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0002559670781893004,
+      "loss": 0.734,
+      "step": 1900
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00025555555555555553,
+      "loss": 0.7416,
+      "step": 1905
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.0002551440329218107,
+      "loss": 0.7071,
+      "step": 1910
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 0.00025473251028806584,
+      "loss": 0.7275,
+      "step": 1915
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.000254320987654321,
+      "loss": 0.7413,
+      "step": 1920
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0002539094650205761,
+      "loss": 0.7703,
+      "step": 1925
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.00025349794238683127,
+      "loss": 0.7331,
+      "step": 1930
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0002530864197530864,
+      "loss": 0.7251,
+      "step": 1935
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 0.0002526748971193416,
+      "loss": 0.7341,
+      "step": 1940
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00025226337448559674,
+      "loss": 0.7079,
+      "step": 1945
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00025185185185185185,
+      "loss": 0.7615,
+      "step": 1950
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.000251440329218107,
+      "loss": 0.7144,
+      "step": 1955
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.00025102880658436216,
+      "loss": 0.7527,
+      "step": 1960
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 0.0002506172839506173,
+      "loss": 0.7327,
+      "step": 1965
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0002502057613168725,
+      "loss": 0.7173,
+      "step": 1970
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0002497942386831276,
+      "loss": 0.7239,
+      "step": 1975
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.00024938271604938274,
+      "loss": 0.7442,
+      "step": 1980
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.0002489711934156379,
+      "loss": 0.7396,
+      "step": 1985
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 0.000248559670781893,
+      "loss": 0.7365,
+      "step": 1990
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00024814814814814816,
+      "loss": 0.7037,
+      "step": 1995
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0002477366255144033,
+      "loss": 0.721,
+      "step": 2000
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0002473251028806585,
+      "loss": 0.7353,
+      "step": 2005
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.0002469135802469136,
+      "loss": 0.7648,
+      "step": 2010
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 0.00024650205761316874,
+      "loss": 0.7445,
+      "step": 2015
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0002460905349794239,
+      "loss": 0.7664,
+      "step": 2020
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.000245679012345679,
+      "loss": 0.7202,
+      "step": 2025
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.00024526748971193417,
+      "loss": 0.7147,
+      "step": 2030
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0002448559670781893,
+      "loss": 0.722,
+      "step": 2035
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 0.0002444444444444445,
+      "loss": 0.7539,
+      "step": 2040
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.0002440329218106996,
+      "loss": 0.7681,
+      "step": 2045
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.00024362139917695475,
+      "loss": 0.7304,
+      "step": 2050
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.0002432098765432099,
+      "loss": 0.7432,
+      "step": 2055
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.000242798353909465,
+      "loss": 0.6945,
+      "step": 2060
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 0.00024238683127572017,
+      "loss": 0.7426,
+      "step": 2065
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00024197530864197533,
+      "loss": 0.7144,
+      "step": 2070
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00024156378600823048,
+      "loss": 0.7761,
+      "step": 2075
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.0002411522633744856,
+      "loss": 0.7402,
+      "step": 2080
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.00024074074074074075,
+      "loss": 0.7376,
+      "step": 2085
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 0.0002403292181069959,
+      "loss": 0.7231,
+      "step": 2090
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00023991769547325104,
+      "loss": 0.7461,
+      "step": 2095
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00023950617283950617,
+      "loss": 0.7215,
+      "step": 2100
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00023909465020576133,
+      "loss": 0.7419,
+      "step": 2105
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00023868312757201646,
+      "loss": 0.7502,
+      "step": 2110
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 0.00023827160493827162,
+      "loss": 0.7294,
+      "step": 2115
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00023786008230452678,
+      "loss": 0.7904,
+      "step": 2120
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00023744855967078194,
+      "loss": 0.7418,
+      "step": 2125
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00023703703703703704,
+      "loss": 0.7371,
+      "step": 2130
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.0002366255144032922,
+      "loss": 0.7384,
+      "step": 2135
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 0.00023621399176954736,
+      "loss": 0.7552,
+      "step": 2140
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00023580246913580246,
+      "loss": 0.7516,
+      "step": 2145
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00023539094650205762,
+      "loss": 0.7303,
+      "step": 2150
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00023497942386831278,
+      "loss": 0.7496,
+      "step": 2155
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00023456790123456794,
+      "loss": 0.7135,
+      "step": 2160
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 0.00023415637860082304,
+      "loss": 0.7283,
+      "step": 2165
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.0002337448559670782,
+      "loss": 0.751,
+      "step": 2170
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00023333333333333336,
+      "loss": 0.7183,
+      "step": 2175
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00023292181069958846,
+      "loss": 0.7242,
+      "step": 2180
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00023251028806584362,
+      "loss": 0.725,
+      "step": 2185
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 0.00023209876543209878,
+      "loss": 0.753,
+      "step": 2190
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0002316872427983539,
+      "loss": 0.7664,
+      "step": 2195
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.00023127572016460904,
+      "loss": 0.7444,
+      "step": 2200
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0002308641975308642,
+      "loss": 0.7226,
+      "step": 2205
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.00023045267489711936,
+      "loss": 0.7653,
+      "step": 2210
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 0.0002300411522633745,
+      "loss": 0.7382,
+      "step": 2215
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00022962962962962965,
+      "loss": 0.7242,
+      "step": 2220
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.0002292181069958848,
+      "loss": 0.7215,
+      "step": 2225
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.0002288065843621399,
+      "loss": 0.724,
+      "step": 2230
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00022839506172839507,
+      "loss": 0.7439,
+      "step": 2235
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 0.00022798353909465023,
+      "loss": 0.699,
+      "step": 2240
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0002275720164609054,
+      "loss": 0.7602,
+      "step": 2245
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0002271604938271605,
+      "loss": 0.7103,
+      "step": 2250
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.00022674897119341565,
+      "loss": 0.7015,
+      "step": 2255
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.0002263374485596708,
+      "loss": 0.7404,
+      "step": 2260
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 0.00022592592592592591,
+      "loss": 0.7127,
+      "step": 2265
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.00022551440329218107,
+      "loss": 0.7299,
+      "step": 2270
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.00022510288065843623,
+      "loss": 0.7174,
+      "step": 2275
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0002246913580246914,
+      "loss": 0.7454,
+      "step": 2280
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.0002242798353909465,
+      "loss": 0.7159,
+      "step": 2285
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 0.00022386831275720165,
+      "loss": 0.7263,
+      "step": 2290
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.0002234567901234568,
+      "loss": 0.7148,
+      "step": 2295
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00022304526748971192,
+      "loss": 0.7447,
+      "step": 2300
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00022263374485596707,
+      "loss": 0.7326,
+      "step": 2305
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00022222222222222223,
+      "loss": 0.7428,
+      "step": 2310
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 0.00022181069958847736,
+      "loss": 0.7587,
+      "step": 2315
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.00022139917695473252,
+      "loss": 0.759,
+      "step": 2320
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.00022098765432098768,
+      "loss": 0.7716,
+      "step": 2325
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0002205761316872428,
+      "loss": 0.74,
+      "step": 2330
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.00022016460905349794,
+      "loss": 0.7255,
+      "step": 2335
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 0.0002197530864197531,
+      "loss": 0.7547,
+      "step": 2340
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00021934156378600826,
+      "loss": 0.7359,
+      "step": 2345
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00021893004115226337,
+      "loss": 0.6987,
+      "step": 2350
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00021851851851851852,
+      "loss": 0.7233,
+      "step": 2355
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00021810699588477368,
+      "loss": 0.6945,
+      "step": 2360
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 0.00021769547325102884,
+      "loss": 0.7566,
+      "step": 2365
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00021728395061728395,
+      "loss": 0.728,
+      "step": 2370
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.0002168724279835391,
+      "loss": 0.7167,
+      "step": 2375
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00021646090534979426,
+      "loss": 0.7576,
+      "step": 2380
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00021604938271604937,
+      "loss": 0.7246,
+      "step": 2385
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 0.00021563786008230453,
+      "loss": 0.755,
+      "step": 2390
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00021522633744855968,
+      "loss": 0.6983,
+      "step": 2395
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00021481481481481484,
+      "loss": 0.7369,
+      "step": 2400
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00021440329218106995,
+      "loss": 0.7074,
+      "step": 2405
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.0002139917695473251,
+      "loss": 0.7317,
+      "step": 2410
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 0.00021358024691358026,
+      "loss": 0.7086,
+      "step": 2415
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.0002131687242798354,
+      "loss": 0.7356,
+      "step": 2420
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00021275720164609055,
+      "loss": 0.7652,
+      "step": 2425
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00021234567901234569,
+      "loss": 0.7488,
+      "step": 2430
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00021193415637860082,
+      "loss": 0.746,
+      "step": 2435
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 0.00021152263374485598,
+      "loss": 0.7454,
+      "step": 2440
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00021111111111111113,
+      "loss": 0.72,
+      "step": 2445
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0002106995884773663,
+      "loss": 0.7299,
+      "step": 2450
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.0002102880658436214,
+      "loss": 0.7436,
+      "step": 2455
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00020987654320987656,
+      "loss": 0.7068,
+      "step": 2460
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 0.00020946502057613171,
+      "loss": 0.711,
+      "step": 2465
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.00020905349794238682,
+      "loss": 0.7473,
+      "step": 2470
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.00020864197530864198,
+      "loss": 0.7016,
+      "step": 2475
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.00020823045267489714,
+      "loss": 0.7049,
+      "step": 2480
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0002078189300411523,
+      "loss": 0.7268,
+      "step": 2485
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 0.0002074074074074074,
+      "loss": 0.7239,
+      "step": 2490
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.00020699588477366256,
+      "loss": 0.698,
+      "step": 2495
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.00020658436213991772,
+      "loss": 0.7573,
+      "step": 2500
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.00020617283950617282,
+      "loss": 0.7309,
+      "step": 2505
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.00020576131687242798,
+      "loss": 0.6628,
+      "step": 2510
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 0.00020534979423868314,
+      "loss": 0.656,
+      "step": 2515
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.0002049382716049383,
+      "loss": 0.6737,
+      "step": 2520
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.00020452674897119343,
+      "loss": 0.6903,
+      "step": 2525
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.00020411522633744856,
+      "loss": 0.6674,
+      "step": 2530
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.00020370370370370372,
+      "loss": 0.6507,
+      "step": 2535
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 0.00020329218106995885,
+      "loss": 0.6948,
+      "step": 2540
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.000202880658436214,
+      "loss": 0.6662,
+      "step": 2545
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.00020246913580246917,
+      "loss": 0.6624,
+      "step": 2550
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.00020205761316872427,
+      "loss": 0.6791,
+      "step": 2555
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.00020164609053497943,
+      "loss": 0.7088,
+      "step": 2560
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 0.0002012345679012346,
+      "loss": 0.6796,
+      "step": 2565
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.00020082304526748975,
+      "loss": 0.6601,
+      "step": 2570
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.00020041152263374485,
+      "loss": 0.6865,
+      "step": 2575
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.0002,
+      "loss": 0.6725,
+      "step": 2580
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.00019958847736625514,
+      "loss": 0.6743,
+      "step": 2585
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 0.0001991769547325103,
+      "loss": 0.6395,
+      "step": 2590
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.00019876543209876543,
+      "loss": 0.6522,
+      "step": 2595
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.0001983539094650206,
+      "loss": 0.6868,
+      "step": 2600
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.00019794238683127572,
+      "loss": 0.6861,
+      "step": 2605
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.00019753086419753085,
+      "loss": 0.6843,
+      "step": 2610
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 0.000197119341563786,
+      "loss": 0.6732,
+      "step": 2615
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.00019670781893004114,
+      "loss": 0.6676,
+      "step": 2620
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.0001962962962962963,
+      "loss": 0.6782,
+      "step": 2625
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.00019588477366255146,
+      "loss": 0.6937,
+      "step": 2630
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.0001954732510288066,
+      "loss": 0.6823,
+      "step": 2635
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 0.00019506172839506175,
+      "loss": 0.6839,
+      "step": 2640
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.00019465020576131688,
+      "loss": 0.6351,
+      "step": 2645
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.00019423868312757204,
+      "loss": 0.7157,
+      "step": 2650
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.00019382716049382717,
+      "loss": 0.6591,
+      "step": 2655
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.00019341563786008233,
+      "loss": 0.6826,
+      "step": 2660
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 0.00019300411522633746,
+      "loss": 0.6732,
+      "step": 2665
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.0001925925925925926,
+      "loss": 0.6795,
+      "step": 2670
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.00019218106995884775,
+      "loss": 0.6845,
+      "step": 2675
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.00019176954732510288,
+      "loss": 0.649,
+      "step": 2680
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.00019135802469135804,
+      "loss": 0.6684,
+      "step": 2685
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 0.00019094650205761317,
+      "loss": 0.6712,
+      "step": 2690
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.00019053497942386833,
+      "loss": 0.6665,
+      "step": 2695
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.00019012345679012346,
+      "loss": 0.6426,
+      "step": 2700
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.0001897119341563786,
+      "loss": 0.6561,
+      "step": 2705
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.00018930041152263375,
+      "loss": 0.6664,
+      "step": 2710
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 0.00018888888888888888,
+      "loss": 0.6746,
+      "step": 2715
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.00018847736625514404,
+      "loss": 0.6497,
+      "step": 2720
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.00018806584362139917,
+      "loss": 0.6811,
+      "step": 2725
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.00018765432098765433,
+      "loss": 0.6608,
+      "step": 2730
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.00018724279835390946,
+      "loss": 0.6655,
+      "step": 2735
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 0.00018683127572016462,
+      "loss": 0.7014,
+      "step": 2740
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.00018641975308641978,
+      "loss": 0.6889,
+      "step": 2745
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.0001860082304526749,
+      "loss": 0.6805,
+      "step": 2750
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.00018559670781893007,
+      "loss": 0.6629,
+      "step": 2755
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.0001851851851851852,
+      "loss": 0.6593,
+      "step": 2760
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 0.00018477366255144033,
+      "loss": 0.6902,
+      "step": 2765
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.0001843621399176955,
+      "loss": 0.6777,
+      "step": 2770
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.00018395061728395062,
+      "loss": 0.7127,
+      "step": 2775
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.00018353909465020578,
+      "loss": 0.6551,
+      "step": 2780
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.00018312757201646091,
+      "loss": 0.6703,
+      "step": 2785
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 0.00018271604938271605,
+      "loss": 0.6776,
+      "step": 2790
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.0001823045267489712,
+      "loss": 0.6574,
+      "step": 2795
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.00018189300411522634,
+      "loss": 0.7154,
+      "step": 2800
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.0001814814814814815,
+      "loss": 0.6749,
+      "step": 2805
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.00018106995884773663,
+      "loss": 0.695,
+      "step": 2810
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 0.00018065843621399178,
+      "loss": 0.6685,
+      "step": 2815
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.00018024691358024692,
+      "loss": 0.642,
+      "step": 2820
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.00017983539094650205,
+      "loss": 0.6873,
+      "step": 2825
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.0001794238683127572,
+      "loss": 0.6506,
+      "step": 2830
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.00017901234567901234,
+      "loss": 0.6695,
+      "step": 2835
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 0.0001786008230452675,
+      "loss": 0.651,
+      "step": 2840
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.00017818930041152265,
+      "loss": 0.6585,
+      "step": 2845
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.00017777777777777779,
+      "loss": 0.6717,
+      "step": 2850
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.00017736625514403294,
+      "loss": 0.6524,
+      "step": 2855
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.00017695473251028808,
+      "loss": 0.6679,
+      "step": 2860
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 0.00017654320987654323,
+      "loss": 0.6719,
+      "step": 2865
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.00017613168724279837,
+      "loss": 0.6662,
+      "step": 2870
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.00017572016460905352,
+      "loss": 0.6624,
+      "step": 2875
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.00017530864197530866,
+      "loss": 0.69,
+      "step": 2880
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.0001748971193415638,
+      "loss": 0.6629,
+      "step": 2885
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 0.00017448559670781895,
+      "loss": 0.6566,
+      "step": 2890
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.00017407407407407408,
+      "loss": 0.6711,
+      "step": 2895
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.00017366255144032924,
+      "loss": 0.6995,
+      "step": 2900
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.00017325102880658437,
+      "loss": 0.6597,
+      "step": 2905
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.0001728395061728395,
+      "loss": 0.6685,
+      "step": 2910
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 0.00017242798353909466,
+      "loss": 0.6649,
+      "step": 2915
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.0001720164609053498,
+      "loss": 0.6824,
+      "step": 2920
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.00017160493827160495,
+      "loss": 0.6484,
+      "step": 2925
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.00017119341563786008,
+      "loss": 0.6693,
+      "step": 2930
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.00017078189300411524,
+      "loss": 0.629,
+      "step": 2935
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 0.00017037037037037037,
+      "loss": 0.6588,
+      "step": 2940
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.00016995884773662553,
+      "loss": 0.6771,
+      "step": 2945
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.00016954732510288066,
+      "loss": 0.6773,
+      "step": 2950
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.00016913580246913582,
+      "loss": 0.6307,
+      "step": 2955
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.00016872427983539098,
+      "loss": 0.6809,
+      "step": 2960
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 0.0001683127572016461,
+      "loss": 0.6748,
+      "step": 2965
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.00016790123456790124,
+      "loss": 0.6708,
+      "step": 2970
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.0001674897119341564,
+      "loss": 0.6754,
+      "step": 2975
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.00016707818930041153,
+      "loss": 0.6676,
+      "step": 2980
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.0001666666666666667,
+      "loss": 0.6662,
+      "step": 2985
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 0.00016625514403292182,
+      "loss": 0.6933,
+      "step": 2990
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 0.00016584362139917695,
+      "loss": 0.6879,
+      "step": 2995
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 0.0001654320987654321,
+      "loss": 0.6979,
+      "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_4e4_s11_1/checkpoint-3000/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43458f4cf6aeb211c730b848f93ab15f483827c314fb4dc1e79dd1f658b7dbc8
+size 5944

save_lora_instruct_r8_4e4_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_4e4_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_4e4_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_4e4_s11_1/checkpoint-3500/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:314c9406340ddc3077b70bbc3d2fcace468611bc56918c2090fee77bedbe7876
+size 564342240