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": 76409.78728011192,
-      "required_pages_per_second_upper_estimate": 0.005286206379171135
+      "required_token_positions_per_second": 76515.6069716057,
+      "required_pages_per_second_upper_estimate": 0.005293527230964187
     },
     {
       "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": 9551.22341001399,
-      "required_pages_per_second_upper_estimate": 0.003998897789547588
+      "required_token_positions_per_second": 9564.450871450712,
+      "required_pages_per_second_upper_estimate": 0.004004435851430338
     },
     {
       "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": 19102.44682002798,
-      "required_pages_per_second_upper_estimate": 0.0039785359072124285
+      "required_token_positions_per_second": 19128.901742901424,
+      "required_pages_per_second_upper_estimate": 0.003984045769983735
     },
     {
       "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": 28653.670230041975,
-      "required_pages_per_second_upper_estimate": 0.001717743778669876
+      "required_token_positions_per_second": 28693.352614352134,
+      "required_pages_per_second_upper_estimate": 0.0017201226770177777
     },
     {
       "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": 28653.670230041975,
-      "required_pages_per_second_upper_estimate": 0.0022276968262101237
+      "required_token_positions_per_second": 28693.352614352134,
+      "required_pages_per_second_upper_estimate": 0.0022307819570458764
     },
     {
       "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": 28653.670230041975,
-      "required_pages_per_second_upper_estimate": 0.3675385157968981
+      "required_token_positions_per_second": 28693.352614352134,
+      "required_pages_per_second_upper_estimate": 0.3680475188152042
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08192259285172554,
-    "producer_p95_seconds": 0.30448849300228176,
-    "gpu_update_seconds": 0.9649025684330108,
-    "producer_batches_per_second": 12.206644897213321,
-    "gpu_batches_per_second": 1.0363740679268654,
-    "producer_headroom": 11.778223013270837,
+    "producer_batch_seconds": 0.1265542446328709,
+    "producer_p95_seconds": 0.30294875600156956,
+    "gpu_update_seconds": 0.9635681257466839,
+    "producer_batches_per_second": 7.901749980026054,
+    "gpu_batches_per_second": 1.0378093393501207,
+    "producer_headroom": 7.6138744183726015,
     "sustainable": true,
-    "buffer_seconds": 60.78886181127968,
+    "buffer_seconds": 60.70479192204109,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 404650.76587338065,
-    "upload_bytes_per_second": 5391885.15921511,
+    "generation_bytes_per_second": 447425.2385677717,
+    "upload_bytes_per_second": 5364158.44416546,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.8286345687014,
+    "estimated_full_shards_per_hour": 96.00704067015518,
     "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": 1.0792735680006444,
-      "bytes": 7390991,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0645652359999076,
       "bytes": 834051,
@@ -562,6 +556,12 @@
       "bytes": 6270013,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.485799860998668,
+      "bytes": 7350955,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -278,3 +278,4 @@
 {"step": 468000, "loss": 0.5927771925926208, "lr": 0.0003, "grad_norm": 0.16164584636110144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68501470947, "tokens_per_second": 152663.77963761735, "data_wait_s": 8.539002010365948e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 468100, "loss": 0.5713925361633301, "lr": 0.0003, "grad_norm": 0.13694000479246665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68515960376, "tokens_per_second": 149767.124380907, "data_wait_s": 1.3064000086160377e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 468200, "loss": 0.5405552983283997, "lr": 0.0003, "grad_norm": 0.168116357562067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68530560082, "tokens_per_second": 151650.48431861398, "data_wait_s": 1.3097000191919506e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 468300, "loss": 0.5770984888076782, "lr": 0.0003, "grad_norm": 0.16690113001715673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68545227487, "tokens_per_second": 152071.72160576063, "data_wait_s": 1.233999864780344e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:92741fca812c3b9f1837048c4dfc09f4bf956dea5073d6559116ef177a14921a
-size 597120
+oid sha256:20df66c31213fc9ca93626f76fa165cbd3303dde969a5314bed1cd55fc076c0a
+size 599238