Upload fine-tuned Llama 3.1 8B QLoRA model

README.md

@@ -1,65 +1,75 @@
-
 ---
-    library_name: peft
-    base_model: meta-llama/Llama-3.1-8B
-    tags:
-    - llama
-    - lora
-    - qlora
-    - fine-tuned
-    license: llama3.1
-    language:
-    - en
-    pipeline_tag: text-generation
-    ---
-
-    # Llama 3.1 8B - Robot Task Planning (QLoRA Fine-tuned)
-
-    This model is a fine-tuned version of [meta-llama/Llama-3.1-8B](https://huggingface.co/meta-llama/Llama-3.1-8B) specialized for **robot task planning** using QLoRA (4-bit quantization + LoRA).
-
-    The model converts natural language robot commands into structured task sequences for construction robots including excavators and dump trucks.
-
-    ## Model Details
-
-    - **Base Model**: meta-llama/Llama-3.1-8B
-    - **Fine-tuning Method**: QLoRA (4-bit quantization + LoRA)
-    - **LoRA Rank**: 16
-    - **LoRA Alpha**: 32
-    - **Target Modules**: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
-
-    ## Usage
-
-    ```python
-    from transformers import AutoTokenizer, AutoModelForCausalLM
-    from peft import PeftModel
-
-    # Load tokenizer and base model
-    tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-8B")
-    base_model = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-3.1-8B")
-
-    # Load LoRA adapter
-    model = PeftModel.from_pretrained(base_model, "YongdongWang/llama-3.1-8b-dart-qlora")
-
-    # Generate robot task sequence
-    command = "Deploy Excavator 1 to Soil Area 1 for excavation"
-    inputs = tokenizer(command, return_tensors="pt")
-    outputs = model.generate(**inputs, max_new_tokens=5120)
-    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
-    print(response)
-
-    Training Details
-
-    Training Data: DART LLM Tasks - Robot command and task planning dataset
-    Domain: Construction robotics (excavators, dump trucks, soil/rock areas)
-    Training Epochs: 5
-    Batch Size: 16 (with gradient accumulation)
-    Learning Rate: 2e-4
-    Optimizer: paged_adamw_8bit
-
-    Capabilities
-
-    Multi-robot coordination: Handle multiple excavators and dump trucks
-    Task dependencies: Generate proper task sequences with dependencies
-    Spatial reasoning: Understand soil areas, rock areas, puddles, and navigation
-    Action planning: Convert commands to structured JSON task definitions
-    
+library_name: peft
+base_model: meta-llama/Llama-3.1-8B
+tags:
+- llama
+- lora
+- qlora
+- fine-tuned
+- robotics
+- task-planning
+- construction
+license: llama3.1
+language:
+- en
+pipeline_tag: text-generation
+---
+
+# Llama 3.1 8B - Robot Task Planning (QLoRA Fine-tuned)
+
+This model is a QLoRA fine-tuned version of [meta-llama/Llama-3.1-8B](https://huggingface.co/meta-llama/Llama-3.1-8B) specialized for **robot task planning** in construction environments.
+
+The model converts natural language commands into structured task sequences for construction robots including excavators and dump trucks.
+
+## Model Details
+
+- **Base Model**: meta-llama/Llama-3.1-8B
+- **Fine-tuning Method**: QLoRA (4-bit quantization + LoRA)
+- **LoRA Rank**: 16
+- **LoRA Alpha**: 32
+- **Target Modules**: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
+
+## Usage
+
+```python
+from transformers import AutoTokenizer, AutoModelForCausalLM
+from peft import PeftModel
+
+# Load tokenizer and base model
+tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-8B")
+base_model = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-3.1-8B")
+
+# Load LoRA adapter
+model = PeftModel.from_pretrained(base_model, "YongdongWang/llama-3.1-8b-dart-qlora")
+
+# Generate robot task sequence
+command = "Deploy Excavator 1 to Soil Area 1 for excavation"
+inputs = tokenizer(command, return_tensors="pt")
+outputs = model.generate(**inputs, max_new_tokens=512)
+response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+print(response)
+```
+
+## Training Details
+
+- **Training Data**: DART LLM Tasks - Robot command and task planning dataset
+- **Domain**: Construction robotics (excavators, dump trucks, soil/rock areas)
+- **Training Epochs**: 5
+- **Batch Size**: 16 (with gradient accumulation)
+- **Learning Rate**: 2e-4
+- **Optimizer**: paged_adamw_8bit
+
+## Capabilities
+
+- **Multi-robot coordination**: Handle multiple excavators and dump trucks
+- **Task dependencies**: Generate proper task sequences with dependencies
+- **Spatial reasoning**: Understand soil areas, rock areas, puddles, and navigation
+- **Action planning**: Convert commands to structured JSON task definitions
+
+## Example Output
+
+The model generates structured task sequences in JSON format for robot execution.
+
+## Limitations
+
+This model is specifically trained for construction robotics scenarios and may not generalize to other domains without additional fine-tuning.

adapter_config.json

@@ -25,12 +25,12 @@
   "revision": null,
   "target_modules": [
     "k_proj",
-    "down_proj",
-    "gate_proj",
+    "up_proj",
     "v_proj",
+    "gate_proj",
     "o_proj",
-    "up_proj",
-    "q_proj"
+    "q_proj",
+    "down_proj"
   ],
   "task_type": "CAUSAL_LM",
   "trainable_token_indices": null,

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:686aa6705c3b268bfc62872bc6a20ee3f78af943123a851082ea5a8beecf764b
+oid sha256:0db982df9cb0f4027019591b4dbed63879790c994db46e7854dc9c28a18915fd
 size 167832240

checkpoint-24/adapter_config.json

@@ -25,12 +25,12 @@
   "revision": null,
   "target_modules": [
     "k_proj",
-    "down_proj",
-    "gate_proj",
+    "up_proj",
     "v_proj",
+    "gate_proj",
     "o_proj",
-    "up_proj",
-    "q_proj"
+    "q_proj",
+    "down_proj"
   ],
   "task_type": "CAUSAL_LM",
   "trainable_token_indices": null,

checkpoint-24/adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ff80ada358f7d45551b795de60e3448a3be3b37ca434a09529d49259e740e104
+oid sha256:147edd45999f9fd86edaa853b1974dff9b4b86f3dc23bbf9aebbc796ef8e4782
 size 167832240

checkpoint-24/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6b5ab0f12eed125e6b567965fc0df61f90b689c01a7cf9d5b32828e25f5b9211
+oid sha256:bb5da80ca3998ac6feb84f67496a67e5dc7c7b0dc77a7c882b36663fd205e22c
 size 85728532

checkpoint-24/trainer_state.json

@@ -1,6 +1,6 @@
 {
   "best_global_step": 24,
-  "best_metric": 0.026073265820741653,
+  "best_metric": 0.02759450115263462,
   "best_model_checkpoint": "./outputs/llama3.1-8b-lora-qlora-dart-llm/checkpoint-24",
   "epoch": 4.0,
   "eval_steps": 500,
@@ -11,62 +11,62 @@
   "log_history": [
     {
       "epoch": 0.8791208791208791,
-      "grad_norm": 0.5106796026229858,
+      "grad_norm": 0.5126284956932068,
       "learning_rate": 0.00019594929736144976,
-      "loss": 0.6176,
+      "loss": 0.6177,
       "step": 5
     },
     {
       "epoch": 1.0,
-      "eval_loss": 0.15218934416770935,
-      "eval_runtime": 2.7083,
-      "eval_samples_per_second": 4.062,
-      "eval_steps_per_second": 4.062,
+      "eval_loss": 0.13462547957897186,
+      "eval_runtime": 1.9882,
+      "eval_samples_per_second": 5.533,
+      "eval_steps_per_second": 5.533,
       "step": 6
     },
     {
       "epoch": 1.7032967032967035,
-      "grad_norm": 0.4140273630619049,
+      "grad_norm": 4.223966598510742,
       "learning_rate": 0.00015406408174555976,
-      "loss": 0.1118,
+      "loss": 0.113,
       "step": 10
     },
     {
       "epoch": 2.0,
-      "eval_loss": 0.04035872593522072,
-      "eval_runtime": 2.7366,
-      "eval_samples_per_second": 4.02,
-      "eval_steps_per_second": 4.02,
+      "eval_loss": 0.041809309273958206,
+      "eval_runtime": 1.9899,
+      "eval_samples_per_second": 5.528,
+      "eval_steps_per_second": 5.528,
       "step": 12
     },
     {
       "epoch": 2.5274725274725274,
-      "grad_norm": 0.19475381076335907,
+      "grad_norm": 0.15816885232925415,
       "learning_rate": 8.57685161726715e-05,
-      "loss": 0.0205,
+      "loss": 0.0253,
       "step": 15
     },
     {
       "epoch": 3.0,
-      "eval_loss": 0.029018037021160126,
-      "eval_runtime": 2.6384,
-      "eval_samples_per_second": 4.169,
-      "eval_steps_per_second": 4.169,
+      "eval_loss": 0.029711483046412468,
+      "eval_runtime": 2.0074,
+      "eval_samples_per_second": 5.48,
+      "eval_steps_per_second": 5.48,
       "step": 18
     },
     {
       "epoch": 3.3516483516483517,
-      "grad_norm": 0.0900561586022377,
+      "grad_norm": 0.10277726501226425,
       "learning_rate": 2.4425042564574184e-05,
-      "loss": 0.0216,
+      "loss": 0.0245,
       "step": 20
     },
     {
       "epoch": 4.0,
-      "eval_loss": 0.026073265820741653,
-      "eval_runtime": 2.7519,
-      "eval_samples_per_second": 3.997,
-      "eval_steps_per_second": 3.997,
+      "eval_loss": 0.02759450115263462,
+      "eval_runtime": 1.9936,
+      "eval_samples_per_second": 5.518,
+      "eval_steps_per_second": 5.518,
       "step": 24
     }
   ],

checkpoint-24/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5c790f6569290a09d32f8a5b1e2904334b6f58a1c2321c61a41455b1cc5658b8
+oid sha256:2c7263e7d5295b74152f3cc3f85217bffc73db5e4dd4a80e6aeb76028cabe1f6
 size 5432

checkpoint-25/adapter_config.json

@@ -25,12 +25,12 @@
   "revision": null,
   "target_modules": [
     "k_proj",
-    "down_proj",
-    "gate_proj",
+    "up_proj",
     "v_proj",
+    "gate_proj",
     "o_proj",
-    "up_proj",
-    "q_proj"
+    "q_proj",
+    "down_proj"
   ],
   "task_type": "CAUSAL_LM",
   "trainable_token_indices": null,

checkpoint-25/adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:686aa6705c3b268bfc62872bc6a20ee3f78af943123a851082ea5a8beecf764b
+oid sha256:0db982df9cb0f4027019591b4dbed63879790c994db46e7854dc9c28a18915fd
 size 167832240

checkpoint-25/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0b2d76bcc892f6aa50acb4b2f920ced51c29d48a2e262a4e75f95ad24696d858
+oid sha256:ca872ede0234294689c0eb7b7851107caef12b43187810d77b7973fa2d241b01
 size 85728532

checkpoint-25/trainer_state.json

@@ -1,6 +1,6 @@
 {
   "best_global_step": 25,
-  "best_metric": 0.02604079246520996,
+  "best_metric": 0.0275871679186821,
   "best_model_checkpoint": "./outputs/llama3.1-8b-lora-qlora-dart-llm/checkpoint-25",
   "epoch": 4.175824175824176,
   "eval_steps": 500,
@@ -11,77 +11,77 @@
   "log_history": [
     {
       "epoch": 0.8791208791208791,
-      "grad_norm": 0.5106796026229858,
+      "grad_norm": 0.5126284956932068,
       "learning_rate": 0.00019594929736144976,
-      "loss": 0.6176,
+      "loss": 0.6177,
       "step": 5
     },
     {
       "epoch": 1.0,
-      "eval_loss": 0.15218934416770935,
-      "eval_runtime": 2.7083,
-      "eval_samples_per_second": 4.062,
-      "eval_steps_per_second": 4.062,
+      "eval_loss": 0.13462547957897186,
+      "eval_runtime": 1.9882,
+      "eval_samples_per_second": 5.533,
+      "eval_steps_per_second": 5.533,
       "step": 6
     },
     {
       "epoch": 1.7032967032967035,
-      "grad_norm": 0.4140273630619049,
+      "grad_norm": 4.223966598510742,
       "learning_rate": 0.00015406408174555976,
-      "loss": 0.1118,
+      "loss": 0.113,
       "step": 10
     },
     {
       "epoch": 2.0,
-      "eval_loss": 0.04035872593522072,
-      "eval_runtime": 2.7366,
-      "eval_samples_per_second": 4.02,
-      "eval_steps_per_second": 4.02,
+      "eval_loss": 0.041809309273958206,
+      "eval_runtime": 1.9899,
+      "eval_samples_per_second": 5.528,
+      "eval_steps_per_second": 5.528,
       "step": 12
     },
     {
       "epoch": 2.5274725274725274,
-      "grad_norm": 0.19475381076335907,
+      "grad_norm": 0.15816885232925415,
       "learning_rate": 8.57685161726715e-05,
-      "loss": 0.0205,
+      "loss": 0.0253,
       "step": 15
     },
     {
       "epoch": 3.0,
-      "eval_loss": 0.029018037021160126,
-      "eval_runtime": 2.6384,
-      "eval_samples_per_second": 4.169,
-      "eval_steps_per_second": 4.169,
+      "eval_loss": 0.029711483046412468,
+      "eval_runtime": 2.0074,
+      "eval_samples_per_second": 5.48,
+      "eval_steps_per_second": 5.48,
       "step": 18
     },
     {
       "epoch": 3.3516483516483517,
-      "grad_norm": 0.0900561586022377,
+      "grad_norm": 0.10277726501226425,
       "learning_rate": 2.4425042564574184e-05,
-      "loss": 0.0216,
+      "loss": 0.0245,
       "step": 20
     },
     {
       "epoch": 4.0,
-      "eval_loss": 0.026073265820741653,
-      "eval_runtime": 2.7519,
-      "eval_samples_per_second": 3.997,
-      "eval_steps_per_second": 3.997,
+      "eval_loss": 0.02759450115263462,
+      "eval_runtime": 1.9936,
+      "eval_samples_per_second": 5.518,
+      "eval_steps_per_second": 5.518,
       "step": 24
     },
     {
       "epoch": 4.175824175824176,
-      "grad_norm": 0.09460670500993729,
+      "grad_norm": 0.10169998556375504,
       "learning_rate": 0.0,
-      "loss": 0.0157,
+      "loss": 0.0178,
       "step": 25
     },
     {
       "epoch": 4.175824175824176,
-      "eval_loss": 0.02604079246520996,
-      "eval_runtime": 2.8313,
-      "eval_samples_per_second": 3.885,
-      "eval_steps_per_second": 3.885,
+      "eval_loss": 0.0275871679186821,
+      "eval_runtime": 1.9861,
+      "eval_samples_per_second": 5.539,
+      "eval_steps_per_second": 5.539,
       "step": 25
     }
   ],

checkpoint-25/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5c790f6569290a09d32f8a5b1e2904334b6f58a1c2321c61a41455b1cc5658b8
+oid sha256:2c7263e7d5295b74152f3cc3f85217bffc73db5e4dd4a80e6aeb76028cabe1f6
 size 5432

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5c790f6569290a09d32f8a5b1e2904334b6f58a1c2321c61a41455b1cc5658b8
+oid sha256:2c7263e7d5295b74152f3cc3f85217bffc73db5e4dd4a80e6aeb76028cabe1f6
 size 5432