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": 76934.24937330576,
-      "required_pages_per_second_upper_estimate": 0.005322489883698
+      "required_token_positions_per_second": 77519.91678204635,
+      "required_pages_per_second_upper_estimate": 0.005363007713970287
     },
     {
       "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": 9616.78117166322,
-      "required_pages_per_second_upper_estimate": 0.004026345455348383
+      "required_token_positions_per_second": 9689.989597755794,
+      "required_pages_per_second_upper_estimate": 0.004056996294587565
     },
     {
       "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": 19233.56234332644,
-      "required_pages_per_second_upper_estimate": 0.004005843812966625
+      "required_token_positions_per_second": 19379.979195511587,
+      "required_pages_per_second_upper_estimate": 0.004036338582004739
     },
     {
       "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": 28850.34351498966,
-      "required_pages_per_second_upper_estimate": 0.001729534041799772
+      "required_token_positions_per_second": 29069.96879326738,
+      "required_pages_per_second_upper_estimate": 0.0017427002418841386
     },
     {
       "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": 28850.34351498966,
-      "required_pages_per_second_upper_estimate": 0.00224298730904045
+      "required_token_positions_per_second": 29069.96879326738,
+      "required_pages_per_second_upper_estimate": 0.0022600622083970364
     },
     {
       "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": 28850.34351498966,
-      "required_pages_per_second_upper_estimate": 0.3700612295248863
+      "required_token_positions_per_second": 29069.96879326738,
+      "required_pages_per_second_upper_estimate": 0.37287834677938175
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.4242286331874823,
-    "producer_p95_seconds": 0.23735901299914985,
-    "gpu_update_seconds": 0.9583248111286798,
-    "producer_batches_per_second": 2.3572194844237755,
-    "gpu_batches_per_second": 1.0434875403280404,
-    "producer_headroom": 2.2589819171992587,
+    "producer_batch_seconds": 0.21814840160945437,
+    "producer_p95_seconds": 0.23167979199934052,
+    "gpu_update_seconds": 0.9510846123234673,
+    "producer_batches_per_second": 4.584035420943744,
+    "gpu_batches_per_second": 1.0514311629509325,
+    "producer_headroom": 4.359805551205323,
     "sustainable": true,
-    "buffer_seconds": 60.374463101106834,
+    "buffer_seconds": 60.86941518870191,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 405737.57314082235,
-    "upload_bytes_per_second": 5919213.996066579,
+    "generation_bytes_per_second": 392329.06402836315,
+    "upload_bytes_per_second": 5917448.422021412,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.06183810871605,
+    "estimated_full_shards_per_hour": 84.18468418729945,
     "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": 1.0640653479995308,
-      "bytes": 6920167,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.9458256010002515,
       "bytes": 5937789,
@@ -562,6 +556,12 @@
       "bytes": 11590114,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.1193175999997038,
+      "bytes": 7119189,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -111,3 +111,4 @@
 {"step": 496100, "loss": 0.568658709526062, "lr": 0.0003, "grad_norm": 0.15724324929820394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72612733913, "tokens_per_second": 153071.8232457807, "data_wait_s": 9.933000910677947e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 496200, "loss": 0.5768790245056152, "lr": 0.0003, "grad_norm": 0.1476545211548444, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72627194811, "tokens_per_second": 150266.67203299157, "data_wait_s": 8.623999747214839e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 496300, "loss": 0.5547523498535156, "lr": 0.0003, "grad_norm": 0.1692655133963257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72641635228, "tokens_per_second": 151173.60173753466, "data_wait_s": 1.1692000043694861e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 496400, "loss": 0.5905017852783203, "lr": 0.0003, "grad_norm": 0.15937914559476943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72656150286, "tokens_per_second": 151870.15509323424, "data_wait_s": 1.6418000086559914e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6831261c5723ec5fe3de5dc1b6ff3c2fbeb5874e4887a31b3c059f55978d7eec
-size 241065
+oid sha256:a9b5d1c393379eb5583470d6e50bc07c90323084ab8a8921392bab0d8c07be8c
+size 243183