Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76754.60610364293,
-      "required_pages_per_second_upper_estimate": 0.005310061745472911
+      "required_token_positions_per_second": 76842.16323262847,
+      "required_pages_per_second_upper_estimate": 0.005316119150816661
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9594.325762955366,
-      "required_pages_per_second_upper_estimate": 0.004016943844644662
+      "required_token_positions_per_second": 9605.27040407856,
+      "required_pages_per_second_upper_estimate": 0.004021526137332847
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19188.651525910733,
-      "required_pages_per_second_upper_estimate": 0.003996490074076834
+      "required_token_positions_per_second": 19210.54080815712,
+      "required_pages_per_second_upper_estimate": 0.004001049034309564
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28782.977288866095,
-      "required_pages_per_second_upper_estimate": 0.0017254955394059356
+      "required_token_positions_per_second": 28815.81121223568,
+      "required_pages_per_second_upper_estimate": 0.001727463882977444
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28782.977288866095,
-      "required_pages_per_second_upper_estimate": 0.002237749881272056
+      "required_token_positions_per_second": 28815.81121223568,
+      "required_pages_per_second_upper_estimate": 0.0022403025743928753
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28782.977288866095,
-      "required_pages_per_second_upper_estimate": 0.3691971279083913
+      "required_token_positions_per_second": 28815.81121223568,
+      "required_pages_per_second_upper_estimate": 0.36961828622305615
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11011270389063554,
-    "producer_p95_seconds": 0.217977508000331,
-    "gpu_update_seconds": 0.9605677592878784,
-    "producer_batches_per_second": 9.081604253340329,
-    "gpu_batches_per_second": 1.0410509725429,
-    "producer_headroom": 8.723496248370386,
+    "producer_batch_seconds": 0.05586629200777793,
+    "producer_p95_seconds": 0.1942886359993281,
+    "gpu_update_seconds": 0.9594732487787883,
+    "producer_batches_per_second": 17.89988137857397,
+    "gpu_batches_per_second": 1.0422385421092186,
+    "producer_headroom": 17.1744573390553,
     "sustainable": true,
-    "buffer_seconds": 60.51576883513634,
+    "buffer_seconds": 60.44681467306366,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 421978.1280506367,
+    "generation_bytes_per_second": 422585.02815912815,
     "upload_bytes_per_second": 5536575.035833069,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.54668313159299,
+    "estimated_full_shards_per_hour": 90.6769097669638,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -126,3 +126,4 @@
 {"step": 772600, "loss": 0.5385637879371643, "lr": 3e-05, "grad_norm": 0.26398638208890207, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113086143598, "tokens_per_second": 151235.95749930415, "data_wait_s": 1.3949998901807703e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 772700, "loss": 0.5686334371566772, "lr": 3e-05, "grad_norm": 0.2466945608683629, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113100834187, "tokens_per_second": 152501.1983588463, "data_wait_s": 1.549300031911116e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 772800, "loss": 0.4973268210887909, "lr": 3e-05, "grad_norm": 0.24788275396210357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113115549331, "tokens_per_second": 153642.2639014537, "data_wait_s": 1.1411000741645694e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 772900, "loss": 0.5543903708457947, "lr": 3e-05, "grad_norm": 0.25232360033513035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113130229347, "tokens_per_second": 153180.95703776844, "data_wait_s": 6.741000106558204e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0c2b0d68c1c7dea575d1a8b1cf5c3c1ee7f80fcd2f1e26c2137f93d4ca5fbf5b
-size 272835
+oid sha256:2d5270aa38d67b495e846588fcf51725f359a6f8274ab89a25d9e42cce73976d
+size 274953