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": 76745.79012835528,
-      "required_pages_per_second_upper_estimate": 0.005309451835846631
+      "required_token_positions_per_second": 76531.83552435572,
+      "required_pages_per_second_upper_estimate": 0.005294649959898864
     },
     {
       "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": 9593.22376604441,
-      "required_pages_per_second_upper_estimate": 0.004016482461550356
+      "required_token_positions_per_second": 9566.479440544465,
+      "required_pages_per_second_upper_estimate": 0.004005285170948553
     },
     {
       "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": 19186.44753208882,
-      "required_pages_per_second_upper_estimate": 0.003996031040286955
+      "required_token_positions_per_second": 19132.95888108893,
+      "required_pages_per_second_upper_estimate": 0.0039848907648742625
     },
     {
       "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": 28779.671298133228,
-      "required_pages_per_second_upper_estimate": 0.0017252973503094557
+      "required_token_positions_per_second": 28699.43832163339,
+      "required_pages_per_second_upper_estimate": 0.0017204875058769522
     },
     {
       "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": 28779.671298133228,
-      "required_pages_per_second_upper_estimate": 0.002237492855034092
+      "required_token_positions_per_second": 28699.43832163339,
+      "required_pages_per_second_upper_estimate": 0.002231255094018797
     },
     {
       "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": 28779.671298133228,
-      "required_pages_per_second_upper_estimate": 0.3691547222089664
+      "required_token_positions_per_second": 28699.43832163339,
+      "required_pages_per_second_upper_estimate": 0.36812557973388477
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06624929804682722,
-    "producer_p95_seconds": 0.20380691700120224,
-    "gpu_update_seconds": 0.9606781020391073,
-    "producer_batches_per_second": 15.09449955670725,
-    "gpu_batches_per_second": 1.0409313982253048,
-    "producer_headroom": 14.500955185367667,
+    "producer_batch_seconds": 0.10201585637497601,
+    "producer_p95_seconds": 0.19928197899935185,
+    "gpu_update_seconds": 0.9633638014148581,
+    "producer_batches_per_second": 9.802397740252614,
+    "gpu_batches_per_second": 1.0380294531840781,
+    "producer_headroom": 9.443275150030173,
     "sustainable": true,
-    "buffer_seconds": 60.522720428463764,
+    "buffer_seconds": 60.69191948913606,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 409541.48961292586,
+    "generation_bytes_per_second": 406539.1932193551,
     "upload_bytes_per_second": 4778470.89282629,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.87807003298599,
+    "estimated_full_shards_per_hour": 87.23384711680879,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -134,3 +134,4 @@
 {"step": 855900, "loss": 0.526910662651062, "lr": 3e-05, "grad_norm": 0.30600752623208666, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125279875021, "tokens_per_second": 153814.19635047676, "data_wait_s": 1.024799894366879e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 856000, "loss": 0.5293590426445007, "lr": 3e-05, "grad_norm": 0.2959618808929612, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125294573760, "tokens_per_second": 153006.3369240086, "data_wait_s": 1.4912999176885933e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 856100, "loss": 0.5236691236495972, "lr": 3e-05, "grad_norm": 0.3054200401908126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125309219881, "tokens_per_second": 152672.06349599964, "data_wait_s": 1.4371000361279584e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 856200, "loss": 0.49121734499931335, "lr": 3e-05, "grad_norm": 0.3282350773532133, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125323702384, "tokens_per_second": 150488.83834897677, "data_wait_s": 2.2682999770040624e-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 sha256:3915761b1eb09dbeec2300ea0918bc39c69d3dfba46a13cdf973222e19431ced
-size 290562
+oid sha256:3539df92395e5bf44f3ffa69aaf0659d9ddb3b592fa48c07778c898fcc63dcfb
+size 292680