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": 75407.33187047612,
-      "required_pages_per_second_upper_estimate": 0.006536116408397531
+      "required_token_positions_per_second": 75582.59067643483,
+      "required_pages_per_second_upper_estimate": 0.006551307397508649
     },
     {
       "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": 9425.916483809515,
-      "required_pages_per_second_upper_estimate": 0.003760139015401913
+      "required_token_positions_per_second": 9447.823834554354,
+      "required_pages_per_second_upper_estimate": 0.0037688781851581117
     },
     {
       "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": 18851.83296761903,
-      "required_pages_per_second_upper_estimate": 0.004361039776130158
+      "required_token_positions_per_second": 18895.64766910871,
+      "required_pages_per_second_upper_estimate": 0.0043711755362073855
     },
     {
       "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": 28277.749451428546,
-      "required_pages_per_second_upper_estimate": 0.0017219665797066374
+      "required_token_positions_per_second": 28343.471503663062,
+      "required_pages_per_second_upper_estimate": 0.0017259687078707603
     },
     {
       "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": 28277.749451428546,
-      "required_pages_per_second_upper_estimate": 0.0021768994305914544
+      "required_token_positions_per_second": 28343.471503663062,
+      "required_pages_per_second_upper_estimate": 0.0021819588961027522
     },
     {
       "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": 28277.749451428546,
-      "required_pages_per_second_upper_estimate": 0.35111500864731177
+      "required_token_positions_per_second": 28343.471503663062,
+      "required_pages_per_second_upper_estimate": 0.35193105657850504
     }
   ],
   "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": 0.4453487782812431,
-    "producer_p95_seconds": 0.2804422049994173,
-    "gpu_update_seconds": 0.9777298595664327,
-    "producer_batches_per_second": 2.2454311065124064,
-    "gpu_batches_per_second": 1.0227773962466922,
-    "producer_headroom": 2.195425040436475,
+    "producer_batch_seconds": 0.08971801463266615,
+    "producer_p95_seconds": 0.29829035100192414,
+    "gpu_update_seconds": 0.9754627268020722,
+    "producer_batches_per_second": 11.146033537348272,
+    "gpu_batches_per_second": 1.0251544959368872,
+    "producer_headroom": 10.872540267369093,
     "sustainable": true,
-    "buffer_seconds": 60.61925129311882,
+    "buffer_seconds": 60.47868906172848,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297512.77284926607,
-    "upload_bytes_per_second": 5846034.93397318,
+    "generation_bytes_per_second": 295703.5443292366,
+    "upload_bytes_per_second": 5848020.060453134,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.839315310559144,
+    "estimated_full_shards_per_hour": 63.45109698684524,
     "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": 1.2004358359990874,
-      "bytes": 6185442,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0889191619990015,
       "bytes": 6163490,
@@ -562,6 +556,12 @@
       "bytes": 7150328,
       "pages": 28,
       "shards": 3
+    },
+    {
+      "seconds": 0.9804463260006742,
+      "bytes": 5040957,
+      "pages": 41,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -167,3 +167,4 @@
 {"step": 131700, "loss": 0.5417986307293177, "lr": 0.0003, "grad_norm": 0.13046259595676132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19273714162, "tokens_per_second": 150209.48819650614, "data_wait_s": 9.391998901264742e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 131800, "loss": 0.584075914695859, "lr": 0.0003, "grad_norm": 0.14925274692364493, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19288224095, "tokens_per_second": 146985.59041550822, "data_wait_s": 1.2844997399952263e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 131900, "loss": 0.5877761356532574, "lr": 0.0003, "grad_norm": 0.13578825804121405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19302738932, "tokens_per_second": 148233.61961941043, "data_wait_s": 1.4311997801996768e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 132000, "loss": 0.5825902055948973, "lr": 0.0003, "grad_norm": 0.14820855424023865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19317194857, "tokens_per_second": 147764.81229459238, "data_wait_s": 1.3799999578623101e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c46bf043471fe71aa340097e2f3a3ce3d8d951c3ad5c8a782cd25f90bf998a7e
-size 360456
+oid sha256:11544e8bba9349413a26dd858be618475baf03824c199d20ab4f3c270381bda3
+size 362574