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": 77525.74843073192,
-      "required_pages_per_second_upper_estimate": 0.005363411160957644
+      "required_token_positions_per_second": 77390.36404913075,
+      "required_pages_per_second_upper_estimate": 0.00535404495530343
     },
     {
       "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": 9690.71855384149,
-      "required_pages_per_second_upper_estimate": 0.004057301493278312
+      "required_token_positions_per_second": 9673.795506141343,
+      "required_pages_per_second_upper_estimate": 0.0040502161665996825
     },
     {
       "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": 19381.43710768298,
-      "required_pages_per_second_upper_estimate": 0.004036642226662311
+      "required_token_positions_per_second": 19347.591012282686,
+      "required_pages_per_second_upper_estimate": 0.004029592977571989
     },
     {
       "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": 29072.15566152447,
-      "required_pages_per_second_upper_estimate": 0.0017428313412969965
+      "required_token_positions_per_second": 29021.38651842403,
+      "required_pages_per_second_upper_estimate": 0.0017397878086881182
     },
     {
       "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": 29072.15566152447,
-      "required_pages_per_second_upper_estimate": 0.0022602322277850083
+      "required_token_positions_per_second": 29021.38651842403,
+      "required_pages_per_second_upper_estimate": 0.00225628515021881
     },
     {
       "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": 29072.15566152447,
-      "required_pages_per_second_upper_estimate": 0.3729063975773075
+      "required_token_positions_per_second": 29021.38651842403,
+      "required_pages_per_second_upper_estimate": 0.37225518552127385
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.22107558220281476,
-    "producer_p95_seconds": 0.21815531899846974,
-    "gpu_update_seconds": 0.9510130697528815,
-    "producer_batches_per_second": 4.523339891434052,
-    "gpu_batches_per_second": 1.0515102597484256,
-    "producer_headroom": 4.301755355688363,
+    "producer_batch_seconds": 0.3695117647108077,
+    "producer_p95_seconds": 0.20886131199949887,
+    "gpu_update_seconds": 0.9526767434921779,
+    "producer_batches_per_second": 2.706273779354856,
+    "gpu_batches_per_second": 1.0496739915517952,
+    "producer_headroom": 2.578204091114053,
     "sustainable": true,
-    "buffer_seconds": 60.864836464184414,
+    "buffer_seconds": 60.01863484000721,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 396325.959814011,
-    "upload_bytes_per_second": 3566933.3916712417,
+    "generation_bytes_per_second": 376664.3339945941,
+    "upload_bytes_per_second": 3650767.8953550556,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.04232497992751,
+    "estimated_full_shards_per_hour": 80.82339777830474,
     "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": 1.2297464679995755,
-      "bytes": 644185,
-      "pages": 19,
-      "shards": 1
-    },
     {
       "seconds": 1.3486207620007917,
       "bytes": 11139901,
@@ -562,6 +556,12 @@
       "bytes": 8272426,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.257975387001352,
+      "bytes": 10890364,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -177,3 +177,4 @@
 {"step": 942700, "loss": 0.5388817191123962, "lr": 3e-05, "grad_norm": 0.35459493587676943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137984743582, "tokens_per_second": 152744.6003907176, "data_wait_s": 1.1482999980216846e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 942800, "loss": 0.5239560604095459, "lr": 3e-05, "grad_norm": 0.33354874236633114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137999347824, "tokens_per_second": 152511.57463859563, "data_wait_s": 9.685998520581052e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 942900, "loss": 0.5169530510902405, "lr": 3e-05, "grad_norm": 0.3398005991344303, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138014046013, "tokens_per_second": 143466.18323986366, "data_wait_s": 1.3903001672588289e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 943000, "loss": 0.5173076391220093, "lr": 3e-05, "grad_norm": 0.33327081484553933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138028788788, "tokens_per_second": 153344.13158151816, "data_wait_s": 1.4615998225053772e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f2865c58523c4ee2fb9f70367cf11e2a89e24dba612ec631f6f5101aad10bb4a
-size 381636
+oid sha256:2f082abd87fb08964a8e6aefa593e46bb0cd9f19be367ed3a534e2c3e4448659
+size 383754