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": 75578.38079308633,
-      "required_pages_per_second_upper_estimate": 0.006550942495489853
+      "required_token_positions_per_second": 75522.24355963922,
+      "required_pages_per_second_upper_estimate": 0.006546076662373175
     },
     {
       "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": 9447.29759913579,
-      "required_pages_per_second_upper_estimate": 0.003768668261981726
+      "required_token_positions_per_second": 9440.280444954902,
+      "required_pages_per_second_upper_estimate": 0.003765869014263165
     },
     {
       "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": 18894.59519827158,
-      "required_pages_per_second_upper_estimate": 0.004370932065602071
+      "required_token_positions_per_second": 18880.560889909804,
+      "required_pages_per_second_upper_estimate": 0.004367685475357964
     },
     {
       "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": 28341.892797407374,
-      "required_pages_per_second_upper_estimate": 0.0017258725729426205
+      "required_token_positions_per_second": 28320.84133486471,
+      "required_pages_per_second_upper_estimate": 0.0017245906493222603
     },
     {
       "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": 28341.892797407374,
-      "required_pages_per_second_upper_estimate": 0.0021818373629250505
+      "required_token_positions_per_second": 28320.84133486471,
+      "required_pages_per_second_upper_estimate": 0.0021802167630643383
     },
     {
       "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": 28341.892797407374,
-      "required_pages_per_second_upper_estimate": 0.3519114543303994
+      "required_token_positions_per_second": 28320.84133486471,
+      "required_pages_per_second_upper_estimate": 0.35165006562033235
     }
   ],
   "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": 0.07607856419531345,
-    "producer_p95_seconds": 0.2966165670004557,
-    "gpu_update_seconds": 0.9755170622383115,
-    "producer_batches_per_second": 13.144306948705552,
-    "gpu_batches_per_second": 1.0250973957395606,
-    "producer_headroom": 12.822495699759864,
+    "producer_batch_seconds": 0.11966796952341952,
+    "producer_p95_seconds": 0.2915029389987467,
+    "gpu_update_seconds": 0.9762421840894819,
+    "producer_batches_per_second": 8.356454981082434,
+    "gpu_batches_per_second": 1.0243359857806968,
+    "producer_headroom": 8.157923861977347,
     "sustainable": true,
-    "buffer_seconds": 60.48205785877531,
+    "buffer_seconds": 60.52701541354788,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 293438.5551414113,
-    "upload_bytes_per_second": 5255390.5287232315,
+    "generation_bytes_per_second": 368988.23363008356,
+    "upload_bytes_per_second": 5162218.129315964,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.96508303338771,
+    "estimated_full_shards_per_hour": 79.1762853305519,
     "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": 1.3779209780004749,
-      "bytes": 11061746,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 0.9865021240002534,
       "bytes": 862370,
@@ -562,6 +556,12 @@
       "bytes": 10591031,
       "pages": 45,
       "shards": 3
+    },
+    {
+      "seconds": 0.8998503880029602,
+      "bytes": 1160065,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -206,3 +206,4 @@
 {"step": 215600, "loss": 0.6187153682112694, "lr": 0.0003, "grad_norm": 0.162110905284232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31557255926, "tokens_per_second": 150366.4233226348, "data_wait_s": 1.19420001283288e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 215700, "loss": 0.614033866673708, "lr": 0.0003, "grad_norm": 0.1589129658866797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31571868086, "tokens_per_second": 147939.52947717562, "data_wait_s": 1.1487001756904647e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 215800, "loss": 0.6022542594000697, "lr": 0.0003, "grad_norm": 0.14141947746065262, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31586368098, "tokens_per_second": 147174.885119373, "data_wait_s": 1.3077999028610066e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 215900, "loss": 0.5656412690877914, "lr": 0.0003, "grad_norm": 0.1673837270927268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31600938353, "tokens_per_second": 147248.75614039652, "data_wait_s": 1.1638003343250602e-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 sha256:0f66dcbe9342660bb373ddf8b8b905e806ea06a268bb89765c11b7f04a3e9528
-size 443841
+oid sha256:7247c6c847346deb93fcff6c6d8273d86c3a63e19eb2caeeab497d1309d8da58
+size 445959