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": 75493.68720149761,
-      "required_pages_per_second_upper_estimate": 0.006543601469624889
+      "required_token_positions_per_second": 75613.12156166608,
+      "required_pages_per_second_upper_estimate": 0.006553953737260673
     },
     {
       "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": 9436.710900187201,
-      "required_pages_per_second_upper_estimate": 0.003764445069485879
+      "required_token_positions_per_second": 9451.64019520826,
+      "required_pages_per_second_upper_estimate": 0.0037704005884826313
     },
     {
       "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": 18873.421800374403,
-      "required_pages_per_second_upper_estimate": 0.0043660339726376346
+      "required_token_positions_per_second": 18903.28039041652,
+      "required_pages_per_second_upper_estimate": 0.004372941231950641
     },
     {
       "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": 28310.132700561604,
-      "required_pages_per_second_upper_estimate": 0.0017239385496770631
+      "required_token_positions_per_second": 28354.92058562478,
+      "required_pages_per_second_upper_estimate": 0.0017266658968935382
     },
     {
       "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": 28310.132700561604,
-      "required_pages_per_second_upper_estimate": 0.0021793923827522882
+      "required_token_positions_per_second": 28354.92058562478,
+      "required_pages_per_second_upper_estimate": 0.0021828402781252526
     },
     {
       "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": 28310.132700561604,
-      "required_pages_per_second_upper_estimate": 0.35151710022178134
+      "required_token_positions_per_second": 28354.92058562478,
+      "required_pages_per_second_upper_estimate": 0.35207321585885715
     }
   ],
   "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": 0.11095200952331652,
-    "producer_p95_seconds": 0.34781270700113964,
-    "gpu_update_seconds": 0.9766114589583513,
-    "producer_batches_per_second": 9.012905708479758,
-    "gpu_batches_per_second": 1.023948665384896,
-    "producer_headroom": 8.80210699341247,
+    "producer_batch_seconds": 0.08928544299993746,
+    "producer_p95_seconds": 0.3281994200006011,
+    "gpu_update_seconds": 0.9750688567971807,
+    "producer_batches_per_second": 11.200034030191242,
+    "gpu_batches_per_second": 1.0255685975703408,
+    "producer_headroom": 10.920804377908095,
     "sustainable": true,
-    "buffer_seconds": 60.54991045541778,
+    "buffer_seconds": 60.454269121425206,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 275206.65068525035,
-    "upload_bytes_per_second": 2833233.3178105312,
+    "generation_bytes_per_second": 273841.32478468394,
+    "upload_bytes_per_second": 2782515.4407372265,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.0529407541097,
+    "estimated_full_shards_per_hour": 58.759973599008454,
     "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": 1.4521252480008116,
-      "bytes": 10591031,
-      "pages": 45,
-      "shards": 3
-    },
     {
       "seconds": 0.8998503880029602,
       "bytes": 1160065,
@@ -562,6 +556,12 @@
       "bytes": 11209339,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 0.8518053779989714,
+      "bytes": 1243189,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -283,3 +283,4 @@
 {"step": 223300, "loss": 0.5461439806967974, "lr": 0.0003, "grad_norm": 0.1377671121957184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32684475373, "tokens_per_second": 137166.45368893736, "data_wait_s": 1.1956002708757296e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 223400, "loss": 0.552930623292923, "lr": 0.0003, "grad_norm": 0.156365045295204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32698972515, "tokens_per_second": 147221.9307159658, "data_wait_s": 1.4793000445934013e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 223500, "loss": 0.6047676410526037, "lr": 0.0003, "grad_norm": 0.1741856513177554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32713424774, "tokens_per_second": 146112.18905580763, "data_wait_s": 1.1620999430306256e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 223600, "loss": 0.5933656673878431, "lr": 0.0003, "grad_norm": 0.1388028229036762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32728172930, "tokens_per_second": 150344.67360773784, "data_wait_s": 1.0833999112946913e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8ad3095d0a3e3eba0c40257942476c5d980bef705d61fc9bc99e7f262b64b616
-size 607710
+oid sha256:9e5c28dfa5ebe51f078935998e6c4a5db38a65f68d9ca8cc382a514170a74668
+size 609828