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": 75608.23823963403,
-      "required_pages_per_second_upper_estimate": 0.005230753356700863
+      "required_token_positions_per_second": 76097.9926286936,
+      "required_pages_per_second_upper_estimate": 0.00526463570172275
     },
     {
       "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": 9451.029779954253,
-      "required_pages_per_second_upper_estimate": 0.003956948808922493
+      "required_token_positions_per_second": 9512.2490785867,
+      "required_pages_per_second_upper_estimate": 0.003982580050855571
     },
     {
       "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": 18902.059559908506,
-      "required_pages_per_second_upper_estimate": 0.003936800525496963
+      "required_token_positions_per_second": 19024.4981571734,
+      "required_pages_per_second_upper_estimate": 0.0039623012563842945
     },
     {
       "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": 28353.089339862763,
-      "required_pages_per_second_upper_estimate": 0.0016997244132640772
+      "required_token_positions_per_second": 28536.747235760096,
+      "required_pages_per_second_upper_estimate": 0.0017107344236937526
     },
     {
       "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": 28353.089339862763,
-      "required_pages_per_second_upper_estimate": 0.0022043279841143008
+      "required_token_positions_per_second": 28536.747235760096,
+      "required_pages_per_second_upper_estimate": 0.002218606577694608
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28353.089339862763,
-      "required_pages_per_second_upper_estimate": 0.3636829868762941
+      "required_token_positions_per_second": 28536.747235760096,
+      "required_pages_per_second_upper_estimate": 0.3660387531683803
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2876716992111028,
-    "producer_p95_seconds": 0.2831828000016685,
-    "gpu_update_seconds": 0.9751318337338483,
-    "producer_batches_per_second": 3.4761848410613645,
-    "gpu_batches_per_second": 1.0255023632762863,
-    "producer_headroom": 3.389738498461974,
+    "producer_batch_seconds": 0.12064255759381126,
+    "producer_p95_seconds": 0.2690781480014266,
+    "gpu_update_seconds": 0.9688560427571652,
+    "producer_batches_per_second": 8.288948940944021,
+    "gpu_batches_per_second": 1.0321450823119247,
+    "producer_headroom": 8.03079826953922,
     "sustainable": true,
-    "buffer_seconds": 60.458173691498594,
+    "buffer_seconds": 60.069074650944245,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 385407.6438210163,
-    "upload_bytes_per_second": 5304296.640790743,
+    "generation_bytes_per_second": 409311.35698327195,
+    "upload_bytes_per_second": 5309535.600740369,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.69950853321902,
+    "estimated_full_shards_per_hour": 87.8286889278757,
     "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": 0.9902176969990251,
-      "bytes": 5564439,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.667742324003484,
       "bytes": 13340513,
@@ -562,6 +556,12 @@
       "bytes": 5814992,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.203305928996997,
+      "bytes": 7121753,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -240,3 +240,4 @@
 {"step": 744000, "loss": 0.5613822340965271, "lr": 5e-05, "grad_norm": 0.23755027711396634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108901249779, "tokens_per_second": 151497.57460553377, "data_wait_s": 1.0911000572377816e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 744100, "loss": 0.5697230100631714, "lr": 5e-05, "grad_norm": 0.24556399169410378, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108915801769, "tokens_per_second": 149881.75720913263, "data_wait_s": 1.5187997632892802e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 744200, "loss": 0.5490539073944092, "lr": 5e-05, "grad_norm": 0.22234718924079527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108930354542, "tokens_per_second": 149898.92008542002, "data_wait_s": 1.2029999197693542e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 744300, "loss": 0.5325649976730347, "lr": 5e-05, "grad_norm": 0.24354972110028025, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108945015345, "tokens_per_second": 150717.29578819821, "data_wait_s": 1.0838000889634714e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a4957848766b4c68861ad909eae1b91143c25c00fb1336e35e68fc63897179fb
-size 515853
+oid sha256:65a11ef5fd3d26890eec84cfed2eff6ec8b2c4037b49659922b9e0898d5df1fe
+size 517971