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": 76986.65086055495,
-      "required_pages_per_second_upper_estimate": 0.005326115140174101
+      "required_token_positions_per_second": 75947.70670969416,
+      "required_pages_per_second_upper_estimate": 0.005254238573134994
     },
     {
       "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": 9623.331357569368,
-      "required_pages_per_second_upper_estimate": 0.004029087881403852
+      "required_token_positions_per_second": 9493.46333871177,
+      "required_pages_per_second_upper_estimate": 0.003974714853860795
     },
     {
       "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": 19246.662715138737,
-      "required_pages_per_second_upper_estimate": 0.004008572274935089
+      "required_token_positions_per_second": 18986.92667742354,
+      "required_pages_per_second_upper_estimate": 0.003954476107978951
     },
     {
       "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": 28869.994072708105,
-      "required_pages_per_second_upper_estimate": 0.0017307120627303282
+      "required_token_positions_per_second": 28480.390016135312,
+      "required_pages_per_second_upper_estimate": 0.0017073558944297384
     },
     {
       "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": 28869.994072708105,
-      "required_pages_per_second_upper_estimate": 0.0022445150534693903
+      "required_token_positions_per_second": 28480.390016135312,
+      "required_pages_per_second_upper_estimate": 0.002214225051757992
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28869.994072708105,
-      "required_pages_per_second_upper_estimate": 0.3703132857801735
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28480.390016135312,
+      "required_pages_per_second_upper_estimate": 0.36531586326670146
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07481276007825954,
-    "producer_p95_seconds": 0.26047275500241085,
-    "gpu_update_seconds": 0.9576725208314711,
-    "producer_batches_per_second": 13.366703740831483,
-    "gpu_batches_per_second": 1.0441982809862596,
-    "producer_headroom": 12.80092486668954,
+    "producer_batch_seconds": 0.13043733486742326,
+    "producer_p95_seconds": 0.2691366709987051,
+    "gpu_update_seconds": 0.9707732227099514,
+    "producer_batches_per_second": 7.666516653505699,
+    "gpu_batches_per_second": 1.0301066990789682,
+    "producer_headroom": 7.442449078683239,
     "sustainable": true,
-    "buffer_seconds": 60.33336881238268,
+    "buffer_seconds": 60.187939808016985,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 121982.5995605819,
-    "upload_bytes_per_second": 4760642.536892661,
+    "generation_bytes_per_second": 180290.973096853,
+    "upload_bytes_per_second": 4759826.7819439145,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 26.17462625611394,
+    "estimated_full_shards_per_hour": 38.68624586753075,
     "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": 0.9220264059986221,
-      "bytes": 6175250,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.8423896300009801,
       "bytes": 1034182,
@@ -562,6 +556,12 @@
       "bytes": 10377609,
       "pages": 27,
       "shards": 3
+    },
+    {
+      "seconds": 1.2949794519990974,
+      "bytes": 7882340,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -275,3 +275,4 @@
 {"step": 712500, "loss": 0.52162104845047, "lr": 8e-05, "grad_norm": 0.22757465109347105, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104289867153, "tokens_per_second": 151846.83034889368, "data_wait_s": 1.735600017127581e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 712600, "loss": 0.5429339408874512, "lr": 8e-05, "grad_norm": 0.22336768670284796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104304513287, "tokens_per_second": 151675.49010240284, "data_wait_s": 1.801699909265153e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 712700, "loss": 0.5827353000640869, "lr": 8e-05, "grad_norm": 0.22072225456351155, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104319318055, "tokens_per_second": 153501.52321709527, "data_wait_s": 1.5165998775046319e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 712800, "loss": 0.5862517952919006, "lr": 8e-05, "grad_norm": 0.24374656692797847, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104334017666, "tokens_per_second": 153232.63427555675, "data_wait_s": 1.5704001270933077e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:965876e6fcc613ab87dfd47481ebe60e5739070195bcc8d315a1dd28138ecf9d
-size 590766
+oid sha256:fc9b39cbc41fb4b81195c50708d352551f231d05c80bede522bc009272128145
+size 592884