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": 74811.88327222725,
-      "required_pages_per_second_upper_estimate": 0.006484504433051994
+      "required_token_positions_per_second": 74470.9775934191,
+      "required_pages_per_second_upper_estimate": 0.006454955592830447
     },
     {
       "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": 9351.485409028406,
-      "required_pages_per_second_upper_estimate": 0.0037304473468279904
+      "required_token_positions_per_second": 9308.872199177387,
+      "required_pages_per_second_upper_estimate": 0.003713448300294155
     },
     {
       "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": 18702.970818056812,
-      "required_pages_per_second_upper_estimate": 0.004326603137713291
+      "required_token_positions_per_second": 18617.744398354775,
+      "required_pages_per_second_upper_estimate": 0.004306887505448983
     },
     {
       "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": 28054.45622708522,
-      "required_pages_per_second_upper_estimate": 0.0017083691938731365
+      "required_token_positions_per_second": 27926.616597532164,
+      "required_pages_per_second_upper_estimate": 0.0017005844311560555
     },
     {
       "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": 28054.45622708522,
-      "required_pages_per_second_upper_estimate": 0.002159709700066299
+      "required_token_positions_per_second": 27926.616597532164,
+      "required_pages_per_second_upper_estimate": 0.002149868251500561
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28054.45622708522,
-      "required_pages_per_second_upper_estimate": 0.34834245411531617
+      "required_token_positions_per_second": 27926.616597532164,
+      "required_pages_per_second_upper_estimate": 0.34675511376798446
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.16149126810152836,
-    "producer_p95_seconds": 0.22038491599960253,
-    "gpu_update_seconds": 0.9855118836096776,
-    "producer_batches_per_second": 6.192285265673358,
-    "gpu_batches_per_second": 1.014701107750478,
-    "producer_headroom": 6.102570716022203,
+    "producer_batch_seconds": 0.27418348916410196,
+    "producer_p95_seconds": 0.21773639300045033,
+    "gpu_update_seconds": 0.9900232598331735,
+    "producer_batches_per_second": 3.6471926265461176,
+    "gpu_batches_per_second": 1.0100772785565741,
+    "producer_headroom": 3.6108055333727016,
     "sustainable": true,
-    "buffer_seconds": 60.11622490019033,
+    "buffer_seconds": 60.39141884982359,
     "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": 529813.1333706198,
-    "upload_bytes_per_second": 5185989.27287372,
+    "generation_bytes_per_second": 500126.72371302644,
+    "upload_bytes_per_second": 5248446.352438826,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.68556500281284,
+    "estimated_full_shards_per_hour": 107.31555255454154,
     "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": 1.3289059700000507,
-      "bytes": 6144064,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.1848053769999751,
       "bytes": 6754320,
@@ -562,6 +556,12 @@
       "bytes": 11573294,
       "pages": 33,
       "shards": 4
+    },
+    {
+      "seconds": 1.4719399390014587,
+      "bytes": 11935199,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -120,3 +120,4 @@
 {"step": 207000, "loss": 0.5928605906665325, "lr": 0.0003, "grad_norm": 0.14640750711352307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30297001443, "tokens_per_second": 147390.04179922843, "data_wait_s": 1.5454999811481684e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 207100, "loss": 0.6216017920523882, "lr": 0.0003, "grad_norm": 0.15469489419349103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30311737803, "tokens_per_second": 149235.87903350283, "data_wait_s": 1.9872999473591335e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 207200, "loss": 0.5991527549922466, "lr": 0.0003, "grad_norm": 0.15763295982660674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30326470394, "tokens_per_second": 148571.34491726456, "data_wait_s": 1.2762999176629819e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 207300, "loss": 0.6095955148339272, "lr": 0.0003, "grad_norm": 0.1524404884898676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30341201833, "tokens_per_second": 147164.4404327976, "data_wait_s": 2.0792000213987194e-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:d693db74f226b9208eb374ea0101dd638bfdff71e37364283c1a65795d1197ea
-size 260127
+oid sha256:e01e3254135567c9c5fb830a851bd41480f4eef7f5cd9800c75a36137ba1f81c
+size 262245