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": 77315.03049624784,
-      "required_pages_per_second_upper_estimate": 0.005348833205317065
+      "required_token_positions_per_second": 77416.87157844639,
+      "required_pages_per_second_upper_estimate": 0.005355878807687428
     },
     {
       "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": 9664.37881203098,
-      "required_pages_per_second_upper_estimate": 0.004046273593418607
+      "required_token_positions_per_second": 9677.108947305798,
+      "required_pages_per_second_upper_estimate": 0.00405160343522272
     },
     {
       "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": 19328.75762406196,
-      "required_pages_per_second_upper_estimate": 0.004025670479475459
+      "required_token_positions_per_second": 19354.217894611596,
+      "required_pages_per_second_upper_estimate": 0.004030973182398462
     },
     {
       "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": 28993.13643609294,
-      "required_pages_per_second_upper_estimate": 0.0017380942593877475
+      "required_token_positions_per_second": 29031.3268419174,
+      "required_pages_per_second_upper_estimate": 0.0017403837159035505
     },
     {
       "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": 28993.13643609294,
-      "required_pages_per_second_upper_estimate": 0.0022540888305765472
+      "required_token_positions_per_second": 29031.3268419174,
+      "required_pages_per_second_upper_estimate": 0.0022570579666475565
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28993.13643609294,
-      "required_pages_per_second_upper_estimate": 0.3718928237977058
+      "required_token_positions_per_second": 29031.3268419174,
+      "required_pages_per_second_upper_estimate": 0.372382689317959
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1287583909379464,
-    "producer_p95_seconds": 0.4906667239993112,
-    "gpu_update_seconds": 0.9536050044444861,
-    "producer_batches_per_second": 7.766484131367705,
-    "gpu_batches_per_second": 1.0486522148471116,
-    "producer_headroom": 7.406158134610932,
+    "producer_batch_seconds": 0.1328796950232345,
+    "producer_p95_seconds": 0.3268605670018587,
+    "gpu_update_seconds": 0.9523505470676574,
+    "producer_batches_per_second": 7.525604268019628,
+    "gpu_batches_per_second": 1.0500335229281466,
+    "producer_headroom": 7.16701334166319,
     "sustainable": true,
-    "buffer_seconds": 60.07711528000263,
+    "buffer_seconds": 60.950435012330075,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386734.7681381555,
-    "upload_bytes_per_second": 4893256.670073868,
+    "generation_bytes_per_second": 386976.3455205019,
+    "upload_bytes_per_second": 4929249.605980382,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.98427851780414,
+    "estimated_full_shards_per_hour": 83.03611540042202,
     "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": 1.3202310080014286,
-      "bytes": 5863529,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.5651336049995734,
       "bytes": 6044120,
@@ -562,6 +556,12 @@
       "bytes": 518541,
       "pages": 17,
       "shards": 1
+    },
+    {
+      "seconds": 0.965982398003689,
+      "bytes": 7277534,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -400,3 +400,4 @@
 {"step": 685000, "loss": 0.5596920847892761, "lr": 0.0001, "grad_norm": 0.2115820071756629, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100266499039, "tokens_per_second": 153899.60702044988, "data_wait_s": 8.930001058615744e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 685100, "loss": 0.5791791677474976, "lr": 0.0001, "grad_norm": 0.20834232906286265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100281063608, "tokens_per_second": 141850.859678147, "data_wait_s": 1.203800638904795e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 685200, "loss": 0.5691226124763489, "lr": 0.0001, "grad_norm": 0.2106518709980132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100295657942, "tokens_per_second": 151739.2991392258, "data_wait_s": 1.6899000911507756e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 685300, "loss": 0.6044577956199646, "lr": 0.0001, "grad_norm": 0.2075308347433876, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100310140405, "tokens_per_second": 150438.1131585047, "data_wait_s": 2.190299710491672e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a6fc69e002b3847d7051183bc0cb9b314910ec6396b3b467e13d57a526b7dffd
-size 857865
+oid sha256:446bc42061698b9c54dfa8e626309c8178b911653985296093fedfc225c8f6c7
+size 859983