Training checkpoint 42000 - 0.4B tokens

README.md

@@ -18,10 +18,10 @@ pipeline_tag: text-generation
 
 ## Current Training Status
 
-- **Training Step**: 40,000
-- **Tokens Processed**: 0.41B tokens
-- **Current Loss**: 4.8085
-- **Spike Rate**: 0.0312
+- **Training Step**: 42,000
+- **Tokens Processed**: 0.43B tokens
+- **Current Loss**: 4.6796
+- **Spike Rate**: 0.0368
 - **Learning Rate**: 2.00e-04
 
 ## Model Architecture
@@ -54,7 +54,7 @@ This represents ongoing training of the first large-scale spiking neural network
 from huggingface_hub import hf_hub_download
 checkpoint = hf_hub_download(
     repo_id="rootxhacker/piking-llm-5b-3epochs-exp",
-    filename="checkpoint_40000.pt"
+    filename="checkpoint_42000.pt"
 )
 
 # Load with custom spiking model code
@@ -65,4 +65,4 @@ checkpoint = hf_hub_download(
 
 **🔬 This is live research in progress! Check back for updates as training continues.**
 
-**Training Progress**: 2.7% complete towards 15B tokens
+**Training Progress**: 2.9% complete towards 15B tokens

checkpoint_42000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8a38978b134d0086b74d7263b61b1fde924e7acc833258143cac7ae261a0181b
+size 999026492

config.json

@@ -4,11 +4,11 @@
   "hidden_size": 768,
   "num_layers": 12,
   "max_seq_length": 1024,
-  "training_step": 40000,
-  "tokens_processed": 409600000,
-  "loss": 4.80854635203208,
-  "spike_rate": 0.031233459513321637,
-  "learning_rate": 0.00019997258051146288,
-  "epoch": 0.08192,
-  "progress_percent": 2.730673657191229
+  "training_step": 42000,
+  "tokens_processed": 430080000,
+  "loss": 4.67963318857356,
+  "spike_rate": 0.036837884952164356,
+  "learning_rate": 0.0001999613788968043,
+  "epoch": 0.086016,
+  "progress_percent": 2.8672073400507903
 }