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": 77183.26355310496,
-      "required_pages_per_second_upper_estimate": 0.0053397172624490095
+      "required_token_positions_per_second": 77115.07785727437,
+      "required_pages_per_second_upper_estimate": 0.005335000017798838
     },
     {
       "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": 9647.90794413812,
-      "required_pages_per_second_upper_estimate": 0.0040393775849827005
+      "required_token_positions_per_second": 9639.384732159297,
+      "required_pages_per_second_upper_estimate": 0.004035809094112072
     },
     {
       "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": 19295.81588827624,
-      "required_pages_per_second_upper_estimate": 0.004018809584643286
+      "required_token_positions_per_second": 19278.769464318593,
+      "required_pages_per_second_upper_estimate": 0.004015259264077338
     },
     {
       "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": 28943.723832414362,
-      "required_pages_per_second_upper_estimate": 0.001735132049245894
+      "required_token_positions_per_second": 28918.15419647789,
+      "required_pages_per_second_upper_estimate": 0.001733599188614076
     },
     {
       "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": 28943.723832414362,
-      "required_pages_per_second_upper_estimate": 0.002250247217983612
+      "required_token_positions_per_second": 28918.15419647789,
+      "required_pages_per_second_upper_estimate": 0.0022482592912584937
     },
     {
       "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": 28943.723832414362,
-      "required_pages_per_second_upper_estimate": 0.3712590119728372
+      "required_token_positions_per_second": 28918.15419647789,
+      "required_pages_per_second_upper_estimate": 0.3709310321375802
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.13216689941401683,
-    "producer_p95_seconds": 0.2738011600013124,
-    "gpu_update_seconds": 0.9552329949001493,
-    "producer_batches_per_second": 7.56619096334756,
-    "gpu_batches_per_second": 1.0468650112997093,
-    "producer_headroom": 7.227475253904935,
+    "producer_batch_seconds": 0.09741915258604195,
+    "producer_p95_seconds": 0.3022902480006451,
+    "gpu_update_seconds": 0.9560776186526942,
+    "producer_batches_per_second": 10.264921973292532,
+    "gpu_batches_per_second": 1.045940183611035,
+    "producer_headroom": 9.81406215588124,
     "sustainable": true,
-    "buffer_seconds": 60.17967867870941,
+    "buffer_seconds": 60.23288997511974,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 358321.66213965975,
-    "upload_bytes_per_second": 5294478.523495377,
+    "generation_bytes_per_second": 388723.2164066178,
+    "upload_bytes_per_second": 5304706.563684017,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.88748739378303,
+    "estimated_full_shards_per_hour": 83.4109532275095,
     "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": [
-    {
-      "seconds": 3.700337080997997,
-      "bytes": 13334949,
-      "pages": 41,
-      "shards": 4
-    },
     {
       "seconds": 1.016492090999236,
       "bytes": 885375,
@@ -562,6 +556,12 @@
       "bytes": 7600081,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.2229772380014765,
+      "bytes": 1028666,
+      "pages": 30,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -300,3 +300,4 @@
 {"step": 634400, "loss": 0.542793869972229, "lr": 0.0001, "grad_norm": 0.17187412109977634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92859536246, "tokens_per_second": 150589.42568573117, "data_wait_s": 1.032799991662614e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 634500, "loss": 0.546806812286377, "lr": 0.0001, "grad_norm": 0.15614027594599827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92874067845, "tokens_per_second": 151064.90740583342, "data_wait_s": 1.0116000339621678e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 634600, "loss": 0.49599912762641907, "lr": 0.0001, "grad_norm": 0.16152464281271273, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92888702732, "tokens_per_second": 152388.11125607832, "data_wait_s": 8.424998668488115e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 634700, "loss": 0.5319393277168274, "lr": 0.0001, "grad_norm": 0.17521285304640075, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92903303170, "tokens_per_second": 151029.26757553697, "data_wait_s": 1.3345001207198948e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b07b1d8c60bbd3be262bf699321adcdc27702f45b0be8b4256b2c8fdf05f4066
-size 644499
+oid sha256:a514126ad7ffde216d0aa05414716fd4d9e8d3900287a489fde5e972c3f39970
+size 646617