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": 76758.72571965239,
-      "required_pages_per_second_upper_estimate": 0.005310346750067272
+      "required_token_positions_per_second": 77435.04767092921,
+      "required_pages_per_second_upper_estimate": 0.005357136271939735
     },
     {
       "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": 9594.840714956548,
-      "required_pages_per_second_upper_estimate": 0.0040171594442941355
+      "required_token_positions_per_second": 9679.380958866152,
+      "required_pages_per_second_upper_estimate": 0.004052554679018043
     },
     {
       "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": 19189.681429913097,
-      "required_pages_per_second_upper_estimate": 0.003996704575920131
+      "required_token_positions_per_second": 19358.761917732303,
+      "required_pages_per_second_upper_estimate": 0.004031919582580557
     },
     {
       "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": 28784.522144869647,
-      "required_pages_per_second_upper_estimate": 0.00172558815116449
+      "required_token_positions_per_second": 29038.142876598453,
+      "required_pages_per_second_upper_estimate": 0.0017407923267751973
     },
     {
       "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": 28784.522144869647,
-      "required_pages_per_second_upper_estimate": 0.002237869987031229
+      "required_token_positions_per_second": 29038.142876598453,
+      "required_pages_per_second_upper_estimate": 0.0022575878833634407
     },
     {
       "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": 28784.522144869647,
-      "required_pages_per_second_upper_estimate": 0.3692169436624677
+      "required_token_positions_per_second": 29038.142876598453,
+      "required_pages_per_second_upper_estimate": 0.3724701180923597
     }
   ],
   "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": 0.08075431436714098,
-    "producer_p95_seconds": 0.2799779619999754,
-    "gpu_update_seconds": 0.9605162059265864,
-    "producer_batches_per_second": 12.38323930847342,
-    "gpu_batches_per_second": 1.041106848411084,
-    "producer_headroom": 11.894302037655855,
+    "producer_batch_seconds": 0.12395200638266601,
+    "producer_p95_seconds": 0.2730057869994198,
+    "gpu_update_seconds": 0.9521270047293984,
+    "producer_batches_per_second": 8.067638670670556,
+    "gpu_batches_per_second": 1.0502800519603028,
+    "producer_headroom": 7.681416642744622,
     "sustainable": true,
-    "buffer_seconds": 60.51252097337494,
+    "buffer_seconds": 60.936128302681496,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 262990.5112046861,
-    "upload_bytes_per_second": 4909967.898780187,
+    "generation_bytes_per_second": 263914.0897381053,
+    "upload_bytes_per_second": 4892826.95439903,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.43164159875332,
+    "estimated_full_shards_per_hour": 56.62982005221719,
     "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": 1.2396269109995046,
-      "bytes": 7721145,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.3206068420004158,
       "bytes": 11244076,
@@ -562,6 +556,12 @@
       "bytes": 16602529,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 1.5457586960001208,
+      "bytes": 7678717,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -277,3 +277,4 @@
 {"step": 912700, "loss": 0.5575767159461975, "lr": 3e-05, "grad_norm": 0.34949986137460615, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133592956144, "tokens_per_second": 153514.05849674967, "data_wait_s": 1.4709999959450215e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 912800, "loss": 0.5384781360626221, "lr": 3e-05, "grad_norm": 0.32282612739048333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133607650798, "tokens_per_second": 120936.92530486477, "data_wait_s": 1.052100196829997e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 912900, "loss": 0.5653138756752014, "lr": 3e-05, "grad_norm": 0.3223837000785098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133622263168, "tokens_per_second": 152633.49246152505, "data_wait_s": 1.792699913494289e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 913000, "loss": 0.5310258269309998, "lr": 3e-05, "grad_norm": 0.3406600564195361, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133636845513, "tokens_per_second": 151504.35586913186, "data_wait_s": 1.4197001291904598e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6f9d5db9baf00f538cabc9fb12fefab4da20701f500273ff270ba059b20c2108
-size 595002
+oid sha256:bc16edd5d52dafcd303592c87981d11bf172cb6c0e8cbad5241976350b9fe655
+size 597120