Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76919.69994852546,
-      "required_pages_per_second_upper_estimate": 0.005321483320732383
+      "required_token_positions_per_second": 75948.0229487074,
+      "required_pages_per_second_upper_estimate": 0.005254260451283705
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9614.962493565683,
-      "required_pages_per_second_upper_estimate": 0.004025584012807262
+      "required_token_positions_per_second": 9493.502868588424,
+      "required_pages_per_second_upper_estimate": 0.003974731404194673
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19229.924987131366,
-      "required_pages_per_second_upper_estimate": 0.004005086247594724
+      "required_token_positions_per_second": 18987.00573717685,
+      "required_pages_per_second_upper_estimate": 0.003954492574040618
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28844.887480697053,
-      "required_pages_per_second_upper_estimate": 0.001729206960874812
+      "required_token_positions_per_second": 28480.508605765273,
+      "required_pages_per_second_upper_estimate": 0.0017073630036969789
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28844.887480697053,
-      "required_pages_per_second_upper_estimate": 0.002242563126372758
+      "required_token_positions_per_second": 28480.508605765273,
+      "required_pages_per_second_upper_estimate": 0.002214234271580101
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28844.887480697053,
-      "required_pages_per_second_upper_estimate": 0.36999124537521394
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28480.508605765273,
+      "required_pages_per_second_upper_estimate": 0.36531738440714295
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11722614149238098,
+    "producer_batch_seconds": 0.14875617208599579,
     "producer_p95_seconds": 0.35381016099563567,
-    "gpu_update_seconds": 0.9585060790582731,
-    "producer_batches_per_second": 8.530520473242687,
-    "gpu_batches_per_second": 1.0432902011247487,
-    "producer_headroom": 8.176555731134172,
+    "gpu_update_seconds": 0.970769180519594,
+    "producer_batches_per_second": 6.722410142564714,
+    "gpu_batches_per_second": 1.030110988345098,
+    "producer_headroom": 6.525908585214155,
     "sustainable": true,
-    "buffer_seconds": 60.385882980671205,
+    "buffer_seconds": 60.18768919221483,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340352.2151685737,
-    "upload_bytes_per_second": 5314844.922139221,
+    "generation_bytes_per_second": 352449.9817351,
+    "upload_bytes_per_second": 5306492.160540054,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.03166238110455,
+    "estimated_full_shards_per_hour": 75.62756146468877,
     "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": 0.9475808069983032,
-      "bytes": 1235313,
-      "pages": 36,
-      "shards": 1
-    },
     {
       "seconds": 1.3809676959972421,
       "bytes": 13424902,
@@ -562,6 +556,12 @@
       "bytes": 10613783,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.0124226469997666,
+      "bytes": 904800,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -395,3 +395,4 @@
 {"step": 479700, "loss": 0.5917872190475464, "lr": 0.0003, "grad_norm": 0.15794939545110334, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70212637420, "tokens_per_second": 142967.04182097228, "data_wait_s": 1.138600055128336e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 479800, "loss": 0.5866878628730774, "lr": 0.0003, "grad_norm": 0.14783031110429193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70227140101, "tokens_per_second": 149996.468889314, "data_wait_s": 9.069000952877104e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 479900, "loss": 0.5699813365936279, "lr": 0.0003, "grad_norm": 0.1999884384445087, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70241720642, "tokens_per_second": 150707.2263292181, "data_wait_s": 6.436996045522392e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 480000, "loss": 0.6134145259857178, "lr": 0.0003, "grad_norm": 0.15646532583825484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70256414402, "tokens_per_second": 152872.12864308245, "data_wait_s": 1.1411000741645694e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a9c25d62a519d04b145942083f15b2a5adce241acf29705d05a2a6eea338ae60
-size 846492
+oid sha256:0b7aada626abeb682eda59ad40b92954acac0bc0ab0c9efa014954d59c08bc1c
+size 848610