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": 75714.37117433018,
-      "required_pages_per_second_upper_estimate": 0.0065627297970716
+      "required_token_positions_per_second": 75577.40725271897,
+      "required_pages_per_second_upper_estimate": 0.0065508581114782265
     },
     {
       "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": 9464.296396791273,
-      "required_pages_per_second_upper_estimate": 0.0037754493365211718
+      "required_token_positions_per_second": 9447.175906589871,
+      "required_pages_per_second_upper_estimate": 0.003768619717005693
     },
     {
       "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": 18928.592793582546,
-      "required_pages_per_second_upper_estimate": 0.004378796810940021
+      "required_token_positions_per_second": 18894.351813179743,
+      "required_pages_per_second_upper_estimate": 0.004370875762744505
     },
     {
       "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": 28392.88919037382,
-      "required_pages_per_second_upper_estimate": 0.001728977985716881
+      "required_token_positions_per_second": 28341.527719769612,
+      "required_pages_per_second_upper_estimate": 0.001725850341629895
     },
     {
       "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": 28392.88919037382,
-      "required_pages_per_second_upper_estimate": 0.0021857632064226583
+      "required_token_positions_per_second": 28341.527719769612,
+      "required_pages_per_second_upper_estimate": 0.002181809258237896
     },
     {
       "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": 28392.88919037382,
-      "required_pages_per_second_upper_estimate": 0.35254465885709446
+      "required_token_positions_per_second": 28341.527719769612,
+      "required_pages_per_second_upper_estimate": 0.35190692128797463
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.49446343407809934,
-    "producer_p95_seconds": 0.2367557610014046,
-    "gpu_update_seconds": 0.9737649386302553,
-    "producer_batches_per_second": 2.022394238037938,
-    "gpu_batches_per_second": 1.0269418833323865,
-    "producer_headroom": 1.9693366010891948,
+    "producer_batch_seconds": 0.08124377494536361,
+    "producer_p95_seconds": 0.21642768499987142,
+    "gpu_update_seconds": 0.9755296282322197,
+    "producer_batches_per_second": 12.30863534680041,
+    "gpu_batches_per_second": 1.0250841912532414,
+    "producer_headroom": 12.00743846391016,
     "sustainable": true,
-    "buffer_seconds": 60.37342619507583,
+    "buffer_seconds": 60.48283695039762,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300966.4475045159,
-    "upload_bytes_per_second": 5727728.743224443,
+    "generation_bytes_per_second": 312252.486873165,
+    "upload_bytes_per_second": 5783967.833540881,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.5803934941445,
+    "estimated_full_shards_per_hour": 67.00211481710637,
     "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": 0.9229088270003558,
-      "bytes": 1208028,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 0.9784717390002697,
       "bytes": 7343882,
@@ -562,6 +556,12 @@
       "bytes": 9866864,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.0568891879993316,
+      "bytes": 6084196,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -147,3 +147,4 @@
 {"step": 369700, "loss": 0.613152289763093, "lr": 0.0003, "grad_norm": 0.1619095578312672, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54113773186, "tokens_per_second": 148900.5519914343, "data_wait_s": 1.0379999366705306e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 369800, "loss": 0.5704388450831175, "lr": 0.0003, "grad_norm": 0.15517637246397203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54128242960, "tokens_per_second": 123540.83844821375, "data_wait_s": 1.2939000953338109e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 369900, "loss": 0.5408785194158554, "lr": 0.0003, "grad_norm": 0.14301080626033003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54142696192, "tokens_per_second": 147024.36971539122, "data_wait_s": 1.1666999853332527e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 370000, "loss": 0.6372381877154112, "lr": 0.0003, "grad_norm": 0.327257267922909, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54157295875, "tokens_per_second": 148294.98975789445, "data_wait_s": 1.0888999895541929e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f3532c8761928b9dd51ecb1a6738807866e7d4b051ab9f7f2fc43ef541e33e05
-size 317313
+oid sha256:db7a99d94d9c74c766fef10fe09fb636c40f236ff2d8984fb877584c14624ca2
+size 319431