Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75292.70744070575,
-      "required_pages_per_second_upper_estimate": 0.006526181053337997
+      "required_token_positions_per_second": 75170.76083983257,
+      "required_pages_per_second_upper_estimate": 0.006515611031045144
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9411.588430088219,
-      "required_pages_per_second_upper_estimate": 0.003754423340548994
+      "required_token_positions_per_second": 9396.345104979071,
+      "required_pages_per_second_upper_estimate": 0.003748342550254935
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18823.176860176438,
-      "required_pages_per_second_upper_estimate": 0.004354410690003581
+      "required_token_positions_per_second": 18792.690209958142,
+      "required_pages_per_second_upper_estimate": 0.004347358140022297
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28234.765290264655,
-      "required_pages_per_second_upper_estimate": 0.001719349069812221
+      "required_token_positions_per_second": 28189.03531493721,
+      "required_pages_per_second_upper_estimate": 0.0017165643542414175
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28234.765290264655,
-      "required_pages_per_second_upper_estimate": 0.0021735903908772845
+      "required_token_positions_per_second": 28189.03531493721,
+      "required_pages_per_second_upper_estimate": 0.002170069970787904
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28234.765290264655,
-      "required_pages_per_second_upper_estimate": 0.35058128922438947
+      "required_token_positions_per_second": 28189.03531493721,
+      "required_pages_per_second_upper_estimate": 0.3500134759792047
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09899956067985727,
-    "producer_p95_seconds": 0.3324647819972597,
-    "gpu_update_seconds": 0.9792183400771186,
-    "producer_batches_per_second": 10.101054925221126,
-    "gpu_batches_per_second": 1.0212227029175585,
-    "producer_headroom": 9.891138236902835,
+    "producer_batch_seconds": 0.12554474412513628,
+    "producer_p95_seconds": 0.3390076570030942,
+    "gpu_update_seconds": 0.9808068878947936,
+    "producer_batches_per_second": 7.965287650777746,
+    "gpu_batches_per_second": 1.0195686962867916,
+    "producer_headroom": 7.812408991946153,
     "sustainable": true,
-    "buffer_seconds": 60.711537084781355,
+    "buffer_seconds": 60.810027049477206,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 389610.7740721739,
-    "upload_bytes_per_second": 5904023.614435943,
+    "generation_bytes_per_second": 435553.9938756718,
+    "upload_bytes_per_second": 5908724.956912391,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.60140244125283,
+    "estimated_full_shards_per_hour": 93.4597479076635,
     "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": 1.0460155590008071,
-      "bytes": 6970566,
-      "pages": 24,
-      "shards": 3
-    },
     {
       "seconds": 2.029252282998641,
       "bytes": 16530913,
@@ -562,6 +556,12 @@
       "bytes": 11201873,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 0.8321514729977935,
+      "bytes": 6044525,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -270,3 +270,4 @@
 {"step": 27000, "loss": 0.6335824616253376, "lr": 0.0003, "grad_norm": 0.1356002792379001, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3951104300, "tokens_per_second": 149903.06616645158, "data_wait_s": 1.081600203178823e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 27100, "loss": 0.6737250126898289, "lr": 0.0003, "grad_norm": 0.13846469064501848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3965767704, "tokens_per_second": 136907.83867136188, "data_wait_s": 1.0435000149300322e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 27200, "loss": 0.7051130346953869, "lr": 0.0003, "grad_norm": 0.14441053903586124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3980423900, "tokens_per_second": 149263.70003024803, "data_wait_s": 1.046399847837165e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 27300, "loss": 0.6365917362272739, "lr": 0.0003, "grad_norm": 0.13074144504816781, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3995103119, "tokens_per_second": 148590.62108850246, "data_wait_s": 1.2453001545509323e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5d8759bb500df0c446a98b1bdbbdd10701bfcd93c6ee075ad112761963d4f739
-size 574402
+oid sha256:563ebcb9aece5675f49ec28b855d3dea44db87946201dff189cf06177d3ffe75
+size 576520