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": 77017.64054899347,
-      "required_pages_per_second_upper_estimate": 0.005328259078726262
+      "required_token_positions_per_second": 77135.36086416413,
+      "required_pages_per_second_upper_estimate": 0.005336403243278532
     },
     {
       "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": 9627.205068624184,
-      "required_pages_per_second_upper_estimate": 0.004030709723330217
+      "required_token_positions_per_second": 9641.920108020517,
+      "required_pages_per_second_upper_estimate": 0.004036870603040497
     },
     {
       "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": 19254.410137248367,
-      "required_pages_per_second_upper_estimate": 0.004010185858647265
+      "required_token_positions_per_second": 19283.840216041033,
+      "required_pages_per_second_upper_estimate": 0.0040163153679364065
     },
     {
       "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": 28881.615205872553,
-      "required_pages_per_second_upper_estimate": 0.0017314087319190942
+      "required_token_positions_per_second": 28925.76032406155,
+      "required_pages_per_second_upper_estimate": 0.0017340551643488318
     },
     {
       "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": 28881.615205872553,
-      "required_pages_per_second_upper_estimate": 0.0022454185454569633
+      "required_token_positions_per_second": 28925.76032406155,
+      "required_pages_per_second_upper_estimate": 0.0022488506342223037
     },
     {
       "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": 28881.615205872553,
-      "required_pages_per_second_upper_estimate": 0.3704623491986064
+      "required_token_positions_per_second": 28925.76032406155,
+      "required_pages_per_second_upper_estimate": 0.3710285953753999
     }
   ],
   "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": 0.10980697546088436,
-    "producer_p95_seconds": 0.2787444019995746,
-    "gpu_update_seconds": 0.9572871808907101,
-    "producer_batches_per_second": 9.106889574207623,
-    "gpu_batches_per_second": 1.0446186055364783,
-    "producer_headroom": 8.717908647176214,
+    "producer_batch_seconds": 0.5421446107031329,
+    "producer_p95_seconds": 0.2843153930007247,
+    "gpu_update_seconds": 0.9558262147737338,
+    "producer_batches_per_second": 1.8445263131972351,
+    "gpu_batches_per_second": 1.046215289498754,
+    "producer_headroom": 1.7630466039938637,
     "sustainable": true,
-    "buffer_seconds": 60.309092396114735,
+    "buffer_seconds": 60.21705153074522,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300212.0414400584,
-    "upload_bytes_per_second": 4799553.7997958725,
+    "generation_bytes_per_second": 290643.9291706353,
+    "upload_bytes_per_second": 4798270.27669151,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.4185155143863,
+    "estimated_full_shards_per_hour": 62.36542135562224,
     "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": 1.4041120760011836,
-      "bytes": 7469317,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.2255340310002794,
       "bytes": 10181330,
@@ -562,6 +556,12 @@
       "bytes": 7719999,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.2099752760004776,
+      "bytes": 6423731,
+      "pages": 22,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -258,3 +258,4 @@
 {"step": 868300, "loss": 0.5199347138404846, "lr": 3e-05, "grad_norm": 0.3343284217020211, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127095415088, "tokens_per_second": 153091.7122123852, "data_wait_s": 1.3117998605594039e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 868400, "loss": 0.5221198201179504, "lr": 3e-05, "grad_norm": 0.2929817347336669, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127110113262, "tokens_per_second": 152888.66464888045, "data_wait_s": 9.778003004612401e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 868500, "loss": 0.5200969576835632, "lr": 3e-05, "grad_norm": 0.3253353400520758, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127124625060, "tokens_per_second": 127434.82107878316, "data_wait_s": 1.3767999917035922e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 868600, "loss": 0.5298730731010437, "lr": 3e-05, "grad_norm": 0.31765275095038803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127139099874, "tokens_per_second": 149360.52081760057, "data_wait_s": 1.553399852127768e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ad91d5d77914aabe9f39fc00811ce20c5437908a4eda141139c25201a7ef7782
-size 554760
+oid sha256:ce64af3e8c17187ac8909cd11146f7c2c5d5a364d0d3a3099d997ef857f53cca
+size 556878