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": 77685.1324169386,
-      "required_pages_per_second_upper_estimate": 0.005374437714945222
+      "required_token_positions_per_second": 77255.12765434662,
+      "required_pages_per_second_upper_estimate": 0.0053446889877205965
     },
     {
       "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": 9710.641552117326,
-      "required_pages_per_second_upper_estimate": 0.004065642836616891
+      "required_token_positions_per_second": 9656.890956793328,
+      "required_pages_per_second_upper_estimate": 0.0040431385847948
     },
     {
       "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": 19421.28310423465,
-      "required_pages_per_second_upper_estimate": 0.004044941096934431
+      "required_token_positions_per_second": 19313.781913586656,
+      "required_pages_per_second_upper_estimate": 0.004022551433919497
     },
     {
       "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": 29131.924656351977,
-      "required_pages_per_second_upper_estimate": 0.0017464144012749349
+      "required_token_positions_per_second": 28970.672870379985,
+      "required_pages_per_second_upper_estimate": 0.0017367476029231347
     },
     {
       "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": 29131.924656351977,
-      "required_pages_per_second_upper_estimate": 0.0022648790042368227
+      "required_token_positions_per_second": 28970.672870379985,
+      "required_pages_per_second_upper_estimate": 0.0022523423871491483
     },
     {
       "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": 29131.924656351977,
-      "required_pages_per_second_upper_estimate": 0.37367305006800805
+      "required_token_positions_per_second": 28970.672870379985,
+      "required_pages_per_second_upper_estimate": 0.3716046852962377
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.05716024815632181,
-    "producer_p95_seconds": 0.2035872010001185,
-    "gpu_update_seconds": 0.9490619080662623,
-    "producer_batches_per_second": 17.49467562256204,
-    "gpu_batches_per_second": 1.0536720434155085,
-    "producer_headroom": 16.603530227349054,
+    "producer_batch_seconds": 0.44007020852345136,
+    "producer_p95_seconds": 0.22791605799829995,
+    "gpu_update_seconds": 0.9543444200865523,
+    "producer_batches_per_second": 2.2723646832519226,
+    "gpu_batches_per_second": 1.0478397305548317,
+    "producer_headroom": 2.168618555863218,
     "sustainable": true,
-    "buffer_seconds": 60.73996211624079,
+    "buffer_seconds": 60.1236984654528,
     "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": 443913.8015535069,
-    "upload_bytes_per_second": 5571290.327227649,
+    "generation_bytes_per_second": 386777.67616881843,
+    "upload_bytes_per_second": 5626484.658701362,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.2535680289641,
+    "estimated_full_shards_per_hour": 82.99348558233656,
     "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": 1.4721885819999443,
-      "bytes": 8185407,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.3622697179998795,
       "bytes": 10825600,
@@ -562,6 +556,12 @@
       "bytes": 5804023,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.138313076000486,
+      "bytes": 10599364,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -152,3 +152,4 @@
 {"step": 572300, "loss": 0.5591154098510742, "lr": 0.0003, "grad_norm": 0.15428245129367443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83768000048, "tokens_per_second": 152664.05838696825, "data_wait_s": 8.916998922359198e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 572400, "loss": 0.5954951047897339, "lr": 0.0003, "grad_norm": 0.16859362712613973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83782520139, "tokens_per_second": 151870.59844540505, "data_wait_s": 1.3560000297729857e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 572500, "loss": 0.5691136717796326, "lr": 0.0003, "grad_norm": 0.14838358952968148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83797088863, "tokens_per_second": 143467.99711045585, "data_wait_s": 7.4670006142696366e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 572600, "loss": 0.6073922514915466, "lr": 0.0003, "grad_norm": 0.17427609695454738, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83811616236, "tokens_per_second": 151883.46306007996, "data_wait_s": 8.850000085658394e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0254cfd8261669e838ce2bed3f303eefc911466c13909dd080d3d878b04ab65c
-size 328686
+oid sha256:b05d664374842a793a31e3f115f93869996e322f9d63e866cc48db752e97461a
+size 330804