Upload folder using huggingface_hub

390M_200kbs_main_9k/blumodel_muon_step00006611.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5329d4e89834a62dbf5eed2157dfa4fa35c926267227eeef605978c92f98a139
+size 4205800024

390M_200kbs_main_9k/blumodel_muon_step00009379.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:13e2d7ac4a10745647251b320c0e8c700fd98d2b5d8c4f9f856a7a09092d435f
+size 4205800024

390M_2Mbs_main/blumodel_muon_step00003005.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5439038ff03645cbdf2b37345920d6bfee94b5d00540e0c907ec6009026d95d3
+size 4205800472

390M_2Mbs_main/blumodel_muon_step00003687.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d9ce088d9df8d7d9b6f6770092eb10e93607e44bdceb7121157d49e904504ac8
+size 4205800472

390M_2Mbs_main/blumodel_muon_step00004415.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3848d530a86dd742b69541bc910c47eaaf553f065eea2ad2489ad2515eef1e1
+size 4205800536

390M_2Mbs_main/blumodel_muon_step00004608.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8f198df2b5eb92b80ba04e3825b2d7b238b2f559349491447454f15f6921187
+size 4205800536

check_target_loss.py

@@ -1,3 +1,4 @@
+
 #!/usr/bin/env python3
 """Find when a training run (CSV log) reaches a target loss, after smoothing.
 Two cases, both reported:
@@ -217,3 +218,4 @@ def main():
 
 if __name__ == "__main__":
     main()
+

train_390M.py

@@ -1,4 +1,5 @@
 
+
 #!/usr/bin/env python3
 # =============================================================================
 # train_muon.py — the MUON arm of train.py. Same model, same data order, same
@@ -451,7 +452,7 @@ class ModelConfig:
     # ---- optimization / data ----
     B: int = 8                           # micro-batch
     T: int = 4096                         # sequence length (<= context_length)
-    global_batch: int = 1048576          # tokens per optimizer step
+    global_batch: int = 262144           # tokens per optimizer step
     grad_clip: float = 1.0
     weight_decay: float = 0.1
 
@@ -464,7 +465,7 @@ class ModelConfig:
     muon_lr: float = 0.001
     muon_min_lr: float = 0.0001
     
-    max_steps: int = 13351
+    max_steps: int = 53406
 
     warmup_steps: int = 500
     stable_steps: int = int(max_steps * 0.99) - warmup_steps           # steps held at max_lr after warmup
@@ -1934,5 +1935,6 @@ if __name__ == "__main__":
 
 
 
+