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": 75515.25755253561,
-      "required_pages_per_second_upper_estimate": 0.006545471132983307
+      "required_token_positions_per_second": 75814.25783127,
+      "required_pages_per_second_upper_estimate": 0.006571387719334727
     },
     {
       "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": 9439.407194066951,
-      "required_pages_per_second_upper_estimate": 0.003765520661427697
+      "required_token_positions_per_second": 9476.78222890875,
+      "required_pages_per_second_upper_estimate": 0.0037804301216326597
     },
     {
       "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": 18878.814388133902,
-      "required_pages_per_second_upper_estimate": 0.004367281452909491
+      "required_token_positions_per_second": 18953.5644578175,
+      "required_pages_per_second_upper_estimate": 0.004384573565974497
     },
     {
       "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": 28318.22158220085,
-      "required_pages_per_second_upper_estimate": 0.0017244311201298113
+      "required_token_positions_per_second": 28430.346686726254,
+      "required_pages_per_second_upper_estimate": 0.001731258950720449
     },
     {
       "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": 28318.22158220085,
-      "required_pages_per_second_upper_estimate": 0.002180015087252335
+      "required_token_positions_per_second": 28430.346686726254,
+      "required_pages_per_second_upper_estimate": 0.002188646788180855
     },
     {
       "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": 28318.22158220085,
-      "required_pages_per_second_upper_estimate": 0.35161753705999543
+      "required_token_positions_per_second": 28430.346686726254,
+      "required_pages_per_second_upper_estimate": 0.3530097555996158
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07736459712486976,
-    "producer_p95_seconds": 0.21685933700064197,
-    "gpu_update_seconds": 0.9763324974255405,
-    "producer_batches_per_second": 12.925808925055957,
-    "gpu_batches_per_second": 1.024241231995112,
-    "producer_headroom": 12.619887309045223,
+    "producer_batch_seconds": 0.07552163090622344,
+    "producer_p95_seconds": 0.20367218399769627,
+    "gpu_update_seconds": 0.9724819856983481,
+    "producer_batches_per_second": 13.241239470076037,
+    "gpu_batches_per_second": 1.0282966828243003,
+    "producer_headroom": 12.876866852966886,
     "sustainable": true,
-    "buffer_seconds": 60.53261484038351,
+    "buffer_seconds": 60.29388311329758,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 445721.2477952919,
-    "upload_bytes_per_second": 5814528.070913043,
+    "generation_bytes_per_second": 427671.724478623,
+    "upload_bytes_per_second": 5826980.028211579,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.64140391725604,
+    "estimated_full_shards_per_hour": 91.7683963848974,
     "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": [
@@ -370,6 +370,12 @@
       "bytes": 1182136,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.007938288999867,
+      "bytes": 6307262,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -85,3 +85,4 @@
 {"step": 93500, "loss": 0.6189540736377239, "lr": 0.0003, "grad_norm": 0.12821118007150253, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13682944329, "tokens_per_second": 148068.26681549716, "data_wait_s": 1.4943001588108018e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 93600, "loss": 0.6001891084015369, "lr": 0.0003, "grad_norm": 0.12907642782609183, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13697585392, "tokens_per_second": 149895.60265239142, "data_wait_s": 1.1176001862622797e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 93700, "loss": 0.620792331174016, "lr": 0.0003, "grad_norm": 0.13391458323998845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13712286885, "tokens_per_second": 150802.41198038854, "data_wait_s": 1.0600000678095967e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 93800, "loss": 0.6461169254034758, "lr": 0.0003, "grad_norm": 0.14760715023430984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13726921760, "tokens_per_second": 148834.92666011828, "data_wait_s": 1.0416999430162832e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:68699eaca8f950c94119127ab6b2ecafcbcf5d51ff0748f77c6dee7788e5a52e
-size 185214
+oid sha256:939e5ebaf6b097d6687a89d45bd22e6ce61b105e4a7e0cc7126bdb1bc911f6ff
+size 187332