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": 76791.09529925407,
-      "required_pages_per_second_upper_estimate": 0.005312586152691889
+      "required_token_positions_per_second": 76812.18742525665,
+      "required_pages_per_second_upper_estimate": 0.0053140453549092226
     },
     {
       "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": 9598.886912406759,
-      "required_pages_per_second_upper_estimate": 0.0040188535026723276
+      "required_token_positions_per_second": 9601.523428157081,
+      "required_pages_per_second_upper_estimate": 0.004019957356760278
     },
     {
       "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": 19197.773824813517,
-      "required_pages_per_second_upper_estimate": 0.003998390008367084
+      "required_token_positions_per_second": 19203.046856314162,
+      "required_pages_per_second_upper_estimate": 0.003999488241769477
     },
     {
       "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": 28796.660737220278,
-      "required_pages_per_second_upper_estimate": 0.0017263158412413528
+      "required_token_positions_per_second": 28804.570284471243,
+      "required_pages_per_second_upper_estimate": 0.001726790006521869
     },
     {
       "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": 28796.660737220278,
-      "required_pages_per_second_upper_estimate": 0.0022388137091944654
+      "required_token_positions_per_second": 28804.570284471243,
+      "required_pages_per_second_upper_estimate": 0.0022394286417028062
     },
     {
       "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": 28796.660737220278,
-      "required_pages_per_second_upper_estimate": 0.3693726444917366
+      "required_token_positions_per_second": 28804.570284471243,
+      "required_pages_per_second_upper_estimate": 0.3694740996712618
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.12849819753111547,
-    "producer_p95_seconds": 0.2744159789981495,
-    "gpu_update_seconds": 0.9601113216666955,
-    "producer_batches_per_second": 7.782210328342177,
-    "gpu_batches_per_second": 1.0415458889330251,
-    "producer_headroom": 7.471788243832816,
+    "producer_batch_seconds": 0.09066282374996604,
+    "producer_p95_seconds": 0.26315247299862676,
+    "gpu_update_seconds": 0.9598476813558035,
+    "producer_batches_per_second": 11.029879267359291,
+    "gpu_batches_per_second": 1.0418319692010722,
+    "producer_headroom": 10.587004040409264,
     "sustainable": true,
-    "buffer_seconds": 60.48701326500182,
+    "buffer_seconds": 60.47040392541562,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 359792.7056954616,
-    "upload_bytes_per_second": 3945216.472421042,
+    "generation_bytes_per_second": 360286.5309243906,
+    "upload_bytes_per_second": 3933088.156979335,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.20313909671675,
+    "estimated_full_shards_per_hour": 77.30910249518193,
     "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": 1.593651556999248,
-      "bytes": 11614889,
-      "pages": 46,
-      "shards": 3
-    },
     {
       "seconds": 1.818591990000641,
       "bytes": 7461958,
@@ -562,6 +556,12 @@
       "bytes": 6041654,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.6828175019982154,
+      "bytes": 10664698,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -221,3 +221,4 @@
 {"step": 822100, "loss": 0.5251942873001099, "lr": 3e-05, "grad_norm": 0.286140653402896, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120332322134, "tokens_per_second": 148917.0294094196, "data_wait_s": 1.5284000255633146e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 822200, "loss": 0.508391261100769, "lr": 3e-05, "grad_norm": 0.29185700279355326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120346889303, "tokens_per_second": 151126.83623996298, "data_wait_s": 1.9220999092794955e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 822300, "loss": 0.5214368104934692, "lr": 3e-05, "grad_norm": 0.28538880079113754, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120361651628, "tokens_per_second": 152933.67619670348, "data_wait_s": 1.2795997463399544e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 822400, "loss": 0.524349570274353, "lr": 3e-05, "grad_norm": 0.27834077863244744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120376319050, "tokens_per_second": 151681.88975819264, "data_wait_s": 1.0267001925967634e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e54f5d29f55b6b033f1b50934b5f209dcef37ef4f30ca1492db82784aad9b5a2
-size 475611
+oid sha256:10b820cb12264f83db818f12b4c69ef69b631ec53e8521cd239728a90549a1f3
+size 477729