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": 77547.759392918,
-      "required_pages_per_second_upper_estimate": 0.005364933930394146
+      "required_token_positions_per_second": 76205.54067619873,
+      "required_pages_per_second_upper_estimate": 0.005272076125195006
     },
     {
       "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": 9693.46992411475,
-      "required_pages_per_second_upper_estimate": 0.004058453434556581
+      "required_token_positions_per_second": 9525.69258452484,
+      "required_pages_per_second_upper_estimate": 0.003988208566059543
     },
     {
       "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": 19386.9398482295,
-      "required_pages_per_second_upper_estimate": 0.004037788302401125
+      "required_token_positions_per_second": 19051.38516904968,
+      "required_pages_per_second_upper_estimate": 0.003967901111899939
     },
     {
       "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": 29080.40977234425,
-      "required_pages_per_second_upper_estimate": 0.001743326162637345
+      "required_token_positions_per_second": 28577.077753574526,
+      "required_pages_per_second_upper_estimate": 0.0017131521766555926
     },
     {
       "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": 29080.40977234425,
-      "required_pages_per_second_upper_estimate": 0.0022608739485952497
+      "required_token_positions_per_second": 28577.077753574526,
+      "required_pages_per_second_upper_estimate": 0.0022217420980594794
     },
     {
       "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": 29080.40977234425,
-      "required_pages_per_second_upper_estimate": 0.37301227244832996
+      "required_token_positions_per_second": 28577.077753574526,
+      "required_pages_per_second_upper_estimate": 0.3665560697473676
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08878922366415054,
-    "producer_p95_seconds": 0.3012438799996744,
-    "gpu_update_seconds": 0.950743136580335,
-    "producer_batches_per_second": 11.262628038989817,
-    "gpu_batches_per_second": 1.0518088025298122,
-    "producer_headroom": 10.707866307926807,
+    "producer_batch_seconds": 0.146011549273112,
+    "producer_p95_seconds": 0.3511884990002727,
+    "gpu_update_seconds": 0.9674887068024892,
+    "producer_batches_per_second": 6.8487732989499195,
+    "gpu_batches_per_second": 1.0336037960638933,
+    "producer_headroom": 6.626110822184476,
     "sustainable": true,
-    "buffer_seconds": 60.84756074114144,
+    "buffer_seconds": 60.95178852855682,
     "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": 425367.8552707219,
-    "upload_bytes_per_second": 4952084.233622985,
+    "generation_bytes_per_second": 424115.6160330468,
+    "upload_bytes_per_second": 4970266.512948696,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.27403968421214,
+    "estimated_full_shards_per_hour": 91.00533829444458,
     "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": 1.0057136780014844,
-      "bytes": 5529362,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.374610462000419,
       "bytes": 11892118,
@@ -562,6 +556,12 @@
       "bytes": 10981344,
       "pages": 45,
       "shards": 3
+    },
+    {
+      "seconds": 1.094142590998672,
+      "bytes": 7571723,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -390,3 +390,4 @@
 {"step": 684000, "loss": 0.516822338104248, "lr": 0.0001, "grad_norm": 0.187172208794905, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100119998790, "tokens_per_second": 151498.97683701292, "data_wait_s": 1.5243000234477222e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 684100, "loss": 0.6152196526527405, "lr": 0.0001, "grad_norm": 0.2438437425076607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100134754013, "tokens_per_second": 154608.36312685738, "data_wait_s": 1.4407996786758304e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 684200, "loss": 0.5925822257995605, "lr": 0.0001, "grad_norm": 0.23003865689516462, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100149531747, "tokens_per_second": 152515.86695014147, "data_wait_s": 1.0559997463133186e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 684300, "loss": 0.5552197694778442, "lr": 0.0001, "grad_norm": 0.20512516525554436, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100164277951, "tokens_per_second": 153968.53071679545, "data_wait_s": 9.00599843589589e-06, "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:a23aa686e8c36f6cd62b06fa45bf0a681d5bfea4ddb653e230e6cf4a33670964
-size 835902
+oid sha256:2908bb54b928fad74afa3ad90b7265879178daff8dc7fc4c156165b293bf0a7d
+size 838020