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": 75953.26423321656,
-      "required_pages_per_second_upper_estimate": 0.0065834364419469485
+      "required_token_positions_per_second": 75849.18114715855,
+      "required_pages_per_second_upper_estimate": 0.006574414783843611
     },
     {
       "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": 9494.15802915207,
-      "required_pages_per_second_upper_estimate": 0.003787361588141084
+      "required_token_positions_per_second": 9481.14764339482,
+      "required_pages_per_second_upper_estimate": 0.003782171550739915
     },
     {
       "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": 18988.31605830414,
-      "required_pages_per_second_upper_estimate": 0.004392612737139801
+      "required_token_positions_per_second": 18962.29528678964,
+      "required_pages_per_second_upper_estimate": 0.004386593289599859
     },
     {
       "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": 28482.47408745621,
-      "required_pages_per_second_upper_estimate": 0.0017344332359335683
+      "required_token_positions_per_second": 28443.442930184458,
+      "required_pages_per_second_upper_estimate": 0.0017320564432363745
     },
     {
       "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": 28482.47408745621,
-      "required_pages_per_second_upper_estimate": 0.0021926596997869276
+      "required_token_positions_per_second": 28443.442930184458,
+      "required_pages_per_second_upper_estimate": 0.0021896549732549897
     },
     {
       "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": 28482.47408745621,
-      "required_pages_per_second_upper_estimate": 0.3536570034574942
+      "required_token_positions_per_second": 28443.442930184458,
+      "required_pages_per_second_upper_estimate": 0.3531723671130593
     }
   ],
   "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": 0.16144477301628513,
-    "producer_p95_seconds": 0.4290135090050171,
-    "gpu_update_seconds": 0.9707021909369975,
-    "producer_batches_per_second": 6.194068605114449,
-    "gpu_batches_per_second": 1.0301820778159798,
-    "producer_headroom": 6.012595965798667,
+    "producer_batch_seconds": 0.4779867407660845,
+    "producer_p95_seconds": 0.4442216749957879,
+    "gpu_update_seconds": 0.9720342248251415,
+    "producer_batches_per_second": 2.092108242160166,
+    "gpu_batches_per_second": 1.0287703606114171,
+    "producer_headroom": 2.0336008134184462,
     "sustainable": true,
-    "buffer_seconds": 60.183535838093846,
+    "buffer_seconds": 60.26612193915877,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 307711.7531926669,
-    "upload_bytes_per_second": 6235660.301789231,
+    "generation_bytes_per_second": 283405.777726052,
+    "upload_bytes_per_second": 6229880.644901793,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.0277790721417,
+    "estimated_full_shards_per_hour": 60.812282551156706,
     "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": 1.0399577659991337,
-      "bytes": 7078524,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 0.9194039719986904,
       "bytes": 5850558,
@@ -562,6 +556,12 @@
       "bytes": 917701,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.0914277900010347,
+      "bytes": 7011453,
+      "pages": 24,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -400,3 +400,4 @@
 {"step": 80000, "loss": 0.5926032345741987, "lr": 0.0003, "grad_norm": 0.15032831750193268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11707359392, "tokens_per_second": 146935.6635780565, "data_wait_s": 1.720400177873671e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 80100, "loss": 0.6002968996763229, "lr": 0.0003, "grad_norm": 0.1438176253410346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11722089836, "tokens_per_second": 131454.71190144258, "data_wait_s": 1.8931001250166446e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 80200, "loss": 0.6369344405829906, "lr": 0.0003, "grad_norm": 0.13777636295534612, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11736720088, "tokens_per_second": 138893.55737173365, "data_wait_s": 1.3495002349372953e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 80300, "loss": 0.6221412345767021, "lr": 0.0003, "grad_norm": 0.1311885616147196, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11751349180, "tokens_per_second": 137864.91168652754, "data_wait_s": 1.2386000889819115e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e4503bdd0f28c743d07ad0002a0636b484d363a15bf6f946fbcbfed9d0a0b61f
-size 857865
+oid sha256:829417267f369edd5c374607b2ad2e867388eca5c0a85d716f6c2a485bf34a21
+size 859983