End of training

README.md

@@ -17,7 +17,7 @@ should probably proofread and complete it, then remove this comment. -->
 
 axolotl version: `0.4.1`
 ```yaml
-adapter: qlora
+adapter: lora
 base_model: katuni4ka/tiny-random-qwen1.5-moe
 bf16: auto
 chat_template: llama3
@@ -41,7 +41,7 @@ early_stopping_patience: null
 eval_max_new_tokens: 128
 eval_table_size: null
 evals_per_epoch: 1
-flash_attention: false
+flash_attention: true
 fp16: null
 fsdp: null
 fsdp_config: null
@@ -53,7 +53,7 @@ hub_repo: null
 hub_strategy: end
 hub_token: null
 learning_rate: 0.0001
-load_in_4bit: true
+load_in_4bit: false
 load_in_8bit: false
 local_rank: null
 logging_steps: 1
@@ -64,11 +64,11 @@ lora_model_dir: null
 lora_r: 32
 lora_target_linear: true
 lr_scheduler: cosine
-max_steps: 100
+max_steps: 1000
 micro_batch_size: 1
 mlflow_experiment_name: /tmp/2b8375abdf26554a_train_data.json
 model_type: AutoModelForCausalLM
-num_epochs: 4
+num_epochs: 8
 optimizer: adamw_bnb_8bit
 output_dir: miner_id_24
 pad_to_sequence_len: true
@@ -76,7 +76,7 @@ resume_from_checkpoint: null
 s2_attention: null
 sample_packing: false
 saves_per_epoch: 1
-sequence_len: 512
+sequence_len: 2048
 strict: false
 tf32: false
 tokenizer_type: AutoTokenizer
@@ -101,7 +101,7 @@ xformers_attention: null
 
 This model is a fine-tuned version of [katuni4ka/tiny-random-qwen1.5-moe](https://huggingface.co/katuni4ka/tiny-random-qwen1.5-moe) on the None dataset.
 It achieves the following results on the evaluation set:
-- Loss: 11.9239
+- Loss: 11.8547
 
 ## Model description
 
@@ -129,17 +129,21 @@ The following hyperparameters were used during training:
 - optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
 - lr_scheduler_type: cosine
 - lr_scheduler_warmup_steps: 10
-- training_steps: 100
+- training_steps: 1000
 
 ### Training results
 
 | Training Loss | Epoch  | Step | Validation Loss |
 |:-------------:|:------:|:----:|:---------------:|
-| 11.9457       | 0.0008 | 1    | 11.9393         |
-| 11.9385       | 0.0212 | 25   | 11.9341         |
-| 11.9232       | 0.0424 | 50   | 11.9279         |
-| 11.9167       | 0.0636 | 75   | 11.9244         |
-| 11.9557       | 0.0848 | 100  | 11.9239         |
+| 11.9466       | 0.0008 | 1    | 11.9395         |
+| 11.9117       | 0.1060 | 125  | 11.8985         |
+| 11.8802       | 0.2121 | 250  | 11.8872         |
+| 11.8782       | 0.3181 | 375  | 11.8751         |
+| 11.8794       | 0.4242 | 500  | 11.8600         |
+| 11.8716       | 0.5302 | 625  | 11.8564         |
+| 11.8596       | 0.6363 | 750  | 11.8553         |
+| 11.8706       | 0.7423 | 875  | 11.8548         |
+| 11.8513       | 0.8484 | 1000 | 11.8547         |
 
 
 ### Framework versions

adapter_config.json

@@ -20,15 +20,15 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
-    "v_proj",
-    "shared_expert_gate",
     "o_proj",
-    "up_proj",
-    "down_proj",
-    "k_proj",
     "gate",
-    "q_proj"
+    "k_proj",
+    "up_proj",
+    "v_proj",
+    "shared_expert_gate",
+    "gate_proj",
+    "q_proj",
+    "down_proj"
   ],
   "task_type": "CAUSAL_LM",
   "use_dora": false,

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:87fa5161451e91cdc546edb9eb904dff6858c39afba74b98bacef94af029191a
+oid sha256:2f12be5ef04a053c67b95380acd55b89af25af55306be8bdf60190bfbe142221
 size 1265130

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9c5e03ed8e28bbaac8b3a4213afadd5528ddf7b77685ee80865719d96ac0939c
+oid sha256:e172d6fea7af81f45f881ef2a1d2fd5983609af1043d4bdd89ebb406fba4d06d
 size 1204200

config.json

@@ -11,7 +11,7 @@
   "hidden_size": 32,
   "initializer_range": 0.02,
   "intermediate_size": 22,
-  "max_position_embeddings": 512,
+  "max_position_embeddings": 2048,
   "max_window_layers": 2,
   "mlp_only_layers": [],
   "model_type": "qwen2_moe",
@@ -23,21 +23,6 @@
   "num_hidden_layers": 4,
   "num_key_value_heads": 2,
   "output_router_logits": false,
-  "quantization_config": {
-    "_load_in_4bit": true,
-    "_load_in_8bit": false,
-    "bnb_4bit_compute_dtype": "bfloat16",
-    "bnb_4bit_quant_storage": "float32",
-    "bnb_4bit_quant_type": "nf4",
-    "bnb_4bit_use_double_quant": true,
-    "llm_int8_enable_fp32_cpu_offload": false,
-    "llm_int8_has_fp16_weight": false,
-    "llm_int8_skip_modules": null,
-    "llm_int8_threshold": 6.0,
-    "load_in_4bit": true,
-    "load_in_8bit": false,
-    "quant_method": "bitsandbytes"
-  },
   "rms_norm_eps": 1e-06,
   "rope_scaling": null,
   "rope_theta": 1000000.0,

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7b7b99c2a3eeb003823b295ecde38550d0c43d5d370de60127d11355c6dfa8ad
+oid sha256:4a1a4bf91134d51cf10b3455a9337aef717da260b938001879d17eecb31311f6
 size 6776