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": 75702.82227620087,
-      "required_pages_per_second_upper_estimate": 0.005237296899815758
+      "required_token_positions_per_second": 75099.02718546317,
+      "required_pages_per_second_upper_estimate": 0.005195524954440894
     },
     {
       "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": 9462.852784525108,
-      "required_pages_per_second_upper_estimate": 0.003961898854043899
+      "required_token_positions_per_second": 9387.378398182896,
+      "required_pages_per_second_upper_estimate": 0.003930299304566825
     },
     {
       "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": 18925.705569050217,
-      "required_pages_per_second_upper_estimate": 0.003941725365613985
+      "required_token_positions_per_second": 18774.756796365793,
+      "required_pages_per_second_upper_estimate": 0.003910286717050652
     },
     {
       "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": 28388.558353575325,
-      "required_pages_per_second_upper_estimate": 0.001701850726478097
+      "required_token_positions_per_second": 28162.135194548693,
+      "required_pages_per_second_upper_estimate": 0.001688277003822596
     },
     {
       "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": 28388.558353575325,
-      "required_pages_per_second_upper_estimate": 0.0022070855439187487
+      "required_token_positions_per_second": 28162.135194548693,
+      "required_pages_per_second_upper_estimate": 0.0021894821392346516
     },
     {
       "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": 28388.558353575325,
-      "required_pages_per_second_upper_estimate": 0.36413794530053906
+      "required_token_positions_per_second": 28162.135194548693,
+      "required_pages_per_second_upper_estimate": 0.36123363213079224
     }
   ],
   "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": 0.09536730350754397,
-    "producer_p95_seconds": 0.33196828899963293,
-    "gpu_update_seconds": 0.9739134920360598,
-    "producer_batches_per_second": 10.485774088400179,
-    "gpu_batches_per_second": 1.0267852413764222,
-    "producer_headroom": 10.21223685913505,
+    "producer_batch_seconds": 0.12041136084366144,
+    "producer_p95_seconds": 0.3402846850003698,
+    "gpu_update_seconds": 0.9817437424045812,
+    "producer_batches_per_second": 8.304864200466685,
+    "gpu_batches_per_second": 1.0185957463306095,
+    "producer_headroom": 8.153248460327994,
     "sustainable": true,
-    "buffer_seconds": 60.382636506235706,
+    "buffer_seconds": 60.86811202908403,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 244640.41001139223,
-    "upload_bytes_per_second": 3864936.90270717,
+    "generation_bytes_per_second": 280528.61719710485,
+    "upload_bytes_per_second": 3876412.9613341535,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.49413705116582,
+    "estimated_full_shards_per_hour": 60.1949108785139,
     "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": 3.5694688390030933,
-      "bytes": 7359997,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.4348496300008264,
       "bytes": 6710244,
@@ -562,6 +556,12 @@
       "bytes": 7862983,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 1.568131148997054,
+      "bytes": 886900,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -366,3 +366,4 @@
 {"step": 879100, "loss": 0.526741623878479, "lr": 3e-05, "grad_norm": 0.3095050575443668, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128675793684, "tokens_per_second": 151964.4515750159, "data_wait_s": 1.0734998795669526e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 879200, "loss": 0.5450117588043213, "lr": 3e-05, "grad_norm": 0.3068843426753608, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128690530526, "tokens_per_second": 153357.47921515413, "data_wait_s": 8.413000614382327e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 879300, "loss": 0.545331597328186, "lr": 3e-05, "grad_norm": 0.31861703193024676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128705281439, "tokens_per_second": 152969.91907331857, "data_wait_s": 9.764000424183905e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 879400, "loss": 0.5143944621086121, "lr": 3e-05, "grad_norm": 0.29641323101373607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128719770269, "tokens_per_second": 150172.49895940014, "data_wait_s": 1.1741998605430126e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e82066422a61b3e2533c24dc6617b7750c54efd460e17b93e32330be3929d29b
-size 785070
+oid sha256:545c6a418fc4e80a3ff01f20753d2f815c0e32cf5790ae921b95289bcfac38c4
+size 787188