Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75579.05859252009,
-      "required_pages_per_second_upper_estimate": 0.006551001245426904
+      "required_token_positions_per_second": 75844.63234184774,
+      "required_pages_per_second_upper_estimate": 0.006574020504928111
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9447.38232406501,
-      "required_pages_per_second_upper_estimate": 0.0037687020600227425
+      "required_token_positions_per_second": 9480.579042730968,
+      "required_pages_per_second_upper_estimate": 0.003781944727433767
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18894.76464813002,
-      "required_pages_per_second_upper_estimate": 0.004370971264844605
+      "required_token_positions_per_second": 18961.158085461935,
+      "required_pages_per_second_upper_estimate": 0.004386330218086755
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28342.14697219503,
-      "required_pages_per_second_upper_estimate": 0.0017258880508536367
+      "required_token_positions_per_second": 28441.7371281929,
+      "required_pages_per_second_upper_estimate": 0.0017319525688447401
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28342.14697219503,
-      "required_pages_per_second_upper_estimate": 0.002181856930003805
+      "required_token_positions_per_second": 28441.7371281929,
+      "required_pages_per_second_upper_estimate": 0.0021895236558957176
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28342.14697219503,
-      "required_pages_per_second_upper_estimate": 0.3519146103305938
+      "required_token_positions_per_second": 28441.7371281929,
+      "required_pages_per_second_upper_estimate": 0.3531511867612762
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.37053423446104716,
-    "producer_p95_seconds": 0.3459888149991457,
-    "gpu_update_seconds": 0.9755083137182224,
-    "producer_batches_per_second": 2.698805959062134,
-    "gpu_batches_per_second": 1.0251065889827486,
-    "producer_headroom": 2.6327076501773923,
+    "producer_batch_seconds": 0.10411343603109913,
+    "producer_p95_seconds": 0.3344242249986564,
+    "gpu_update_seconds": 0.9720925228787763,
+    "producer_batches_per_second": 9.604908243555574,
+    "gpu_batches_per_second": 1.0287086634907734,
+    "producer_headroom": 9.336859486497096,
     "sustainable": true,
-    "buffer_seconds": 60.48151545052979,
+    "buffer_seconds": 60.269736418484136,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382187.6634426139,
+    "generation_bytes_per_second": 397170.67908022343,
     "upload_bytes_per_second": 5322141.558102892,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.00857570132077,
+    "estimated_full_shards_per_hour": 85.22358207039859,
     "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": [

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -309,3 +309,4 @@
 {"step": 385900, "loss": 0.5632327254861593, "lr": 0.0003, "grad_norm": 0.1679229761758304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56484749011, "tokens_per_second": 150653.73762363667, "data_wait_s": 8.01799978944473e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 386000, "loss": 0.5998760890215635, "lr": 0.0003, "grad_norm": 0.16240458667847157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56499428794, "tokens_per_second": 149967.0774129678, "data_wait_s": 1.3215001672506332e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 386100, "loss": 0.5972938779741526, "lr": 0.0003, "grad_norm": 0.16952418522300677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56514234255, "tokens_per_second": 151120.4774845238, "data_wait_s": 7.779999577905983e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 386200, "loss": 0.6052970960736275, "lr": 0.0003, "grad_norm": 0.1709050477368543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56528815054, "tokens_per_second": 148483.81847558808, "data_wait_s": 1.5281002561096102e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cce64d39ff708d3058dbe3bf401a532d6bf0172b9b89fdcb1bb827dd698e9a32
-size 663561
+oid sha256:135c80a0a22ffc88bbb8556eff43828ced5b28c4c420e499231eb06d3a62a0c1
+size 665679