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": 77046.42685778001,
-      "required_pages_per_second_upper_estimate": 0.005330250582361553
+      "required_token_positions_per_second": 76904.95070738991,
+      "required_pages_per_second_upper_estimate": 0.0053204629340076704
     },
     {
       "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": 9630.803357222501,
-      "required_pages_per_second_upper_estimate": 0.004032216251625522
+      "required_token_positions_per_second": 9613.11883842374,
+      "required_pages_per_second_upper_estimate": 0.004024812112899227
     },
     {
       "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": 19261.606714445003,
-      "required_pages_per_second_upper_estimate": 0.00401168471589082
+      "required_token_positions_per_second": 19226.23767684748,
+      "required_pages_per_second_upper_estimate": 0.004004318278103502
     },
     {
       "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": 28892.410071667502,
-      "required_pages_per_second_upper_estimate": 0.0017320558676406955
+      "required_token_positions_per_second": 28839.35651527122,
+      "required_pages_per_second_upper_estimate": 0.0017288753879428286
     },
     {
       "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": 28892.410071667502,
-      "required_pages_per_second_upper_estimate": 0.002246257798790924
+      "required_token_positions_per_second": 28839.35651527122,
+      "required_pages_per_second_upper_estimate": 0.0022421331181390457
     },
     {
       "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": 28892.410071667502,
-      "required_pages_per_second_upper_estimate": 0.37060081414640017
+      "required_token_positions_per_second": 28839.35651527122,
+      "required_pages_per_second_upper_estimate": 0.36992030008941934
     }
   ],
   "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": 0.07940675292167043,
-    "producer_p95_seconds": 0.27943740200134926,
-    "gpu_update_seconds": 0.9569295164861378,
-    "producer_batches_per_second": 12.593387378356027,
-    "gpu_batches_per_second": 1.045009044837511,
-    "producer_headroom": 12.050984094892863,
+    "producer_batch_seconds": 0.4650172233984904,
+    "producer_p95_seconds": 0.289552864000143,
+    "gpu_update_seconds": 0.9586899064602789,
+    "producer_batches_per_second": 2.1504579823768446,
+    "gpu_batches_per_second": 1.043090151738687,
+    "producer_headroom": 2.061622361971617,
     "sustainable": true,
-    "buffer_seconds": 60.28655953862668,
+    "buffer_seconds": 60.39746410699757,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 416665.18216713704,
-    "upload_bytes_per_second": 4997680.1440797625,
+    "generation_bytes_per_second": 364473.81704120466,
+    "upload_bytes_per_second": 4967100.541288679,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.40664862404427,
+    "estimated_full_shards_per_hour": 78.20759662081818,
     "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": 1.5982389480013808,
-      "bytes": 6084209,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.98295405799945,
       "bytes": 6764576,
@@ -562,6 +556,12 @@
       "bytes": 5876732,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.0857899119946524,
+      "bytes": 1135012,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -314,3 +314,4 @@
 {"step": 716400, "loss": 0.5153571367263794, "lr": 8e-05, "grad_norm": 0.21063414181555512, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104861215189, "tokens_per_second": 152426.6303756822, "data_wait_s": 1.6742000298108906e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 716500, "loss": 0.6072710156440735, "lr": 8e-05, "grad_norm": 0.24949918891367054, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104875824951, "tokens_per_second": 152561.70198300277, "data_wait_s": 1.3975004549138248e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 716600, "loss": 0.5833146572113037, "lr": 8e-05, "grad_norm": 0.22927128309571412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104890590349, "tokens_per_second": 153805.43548895695, "data_wait_s": 1.6591002349741757e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 716700, "loss": 0.5062534809112549, "lr": 8e-05, "grad_norm": 0.20410455130525226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104905129655, "tokens_per_second": 151253.71473505115, "data_wait_s": 1.3720004062633961e-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 sha256:8148f70d1d78d57b3e3aeeb1f4c04c2c27179eb5d48cf200ab57e8406527874c
-size 674151
+oid sha256:55fb67175c72f2f66bc2a3b2ee3ae833cf8a05a2e2f8fc9e0be5cfd01138a10f
+size 676269