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": 75210.2159865413,
-      "required_pages_per_second_upper_estimate": 0.006519030903163701
+      "required_token_positions_per_second": 75158.8540395207,
+      "required_pages_per_second_upper_estimate": 0.00651457897977162
     },
     {
       "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": 9401.276998317662,
-      "required_pages_per_second_upper_estimate": 0.0037503099562460757
+      "required_token_positions_per_second": 9394.856754940087,
+      "required_pages_per_second_upper_estimate": 0.003747748825171568
     },
     {
       "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": 18802.553996635324,
-      "required_pages_per_second_upper_estimate": 0.004349639953473346
+      "required_token_positions_per_second": 18789.713509880174,
+      "required_pages_per_second_upper_estimate": 0.00434666953284739
     },
     {
       "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": 28203.830994952983,
-      "required_pages_per_second_upper_estimate": 0.0017174653335274444
+      "required_token_positions_per_second": 28184.57026482026,
+      "required_pages_per_second_upper_estimate": 0.0017162924561156029
     },
     {
       "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": 28203.830994952983,
-      "required_pages_per_second_upper_estimate": 0.002171208983192582
+      "required_token_positions_per_second": 28184.57026482026,
+      "required_pages_per_second_upper_estimate": 0.0021697262388698513
     },
     {
       "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": 28203.830994952983,
-      "required_pages_per_second_upper_estimate": 0.350197188807045
+      "required_token_positions_per_second": 28184.57026482026,
+      "required_pages_per_second_upper_estimate": 0.34995803500031364
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.08898600646095645,
-    "producer_p95_seconds": 0.20540237800014438,
-    "gpu_update_seconds": 0.9802923583305952,
-    "producer_batches_per_second": 11.237721971923312,
-    "gpu_batches_per_second": 1.0201038409632879,
-    "producer_headroom": 11.01625297412025,
+    "producer_batch_seconds": 0.07499485353901036,
+    "producer_p95_seconds": 0.1984948199997234,
+    "gpu_update_seconds": 0.9809622690791918,
+    "producer_batches_per_second": 13.334248322517572,
+    "gpu_batches_per_second": 1.0194071999717975,
+    "producer_headroom": 13.080394490922245,
     "sustainable": true,
-    "buffer_seconds": 60.77812621649691,
+    "buffer_seconds": 60.81966068290989,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 554814.518019056,
-    "upload_bytes_per_second": 5417461.619547132,
+    "generation_bytes_per_second": 602341.5674210325,
+    "upload_bytes_per_second": 5363557.320342382,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 119.0502801459194,
+    "estimated_full_shards_per_hour": 64.62423928724876,
     "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": [
@@ -538,6 +538,12 @@
       "bytes": 7396060,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.9060076790001403,
+      "bytes": 1026778,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -72,3 +72,4 @@
 {"step": 322200, "loss": 0.6169965155422688, "lr": 0.0003, "grad_norm": 0.15325261118740036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47160713306, "tokens_per_second": 147730.26352134856, "data_wait_s": 1.538400010758778e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 322300, "loss": 0.5765353012830019, "lr": 0.0003, "grad_norm": 0.157324963402977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47175254349, "tokens_per_second": 147989.39596911386, "data_wait_s": 1.6527000298083294e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 322400, "loss": 0.5849626809358597, "lr": 0.0003, "grad_norm": 0.1463768283313687, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47189920207, "tokens_per_second": 149146.89040614478, "data_wait_s": 1.9225999494665302e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 322500, "loss": 0.5936872847378254, "lr": 0.0003, "grad_norm": 0.1444422420849158, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47204580366, "tokens_per_second": 149559.785035013, "data_wait_s": 1.221999991685152e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:59352f0df98085624335a23e58d5b2816d441fb4bbaae5be695636d0b5762b78
-size 157680
+oid sha256:a1c780e6ba7973d69f6ac091b85e083bfa64badc422720c6e86dd82df3d7257d
+size 159798