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": 77005.97742372737,
-      "required_pages_per_second_upper_estimate": 0.005327452196658173
+      "required_token_positions_per_second": 77720.21097105795,
+      "required_pages_per_second_upper_estimate": 0.005376864530712657
     },
     {
       "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": 9625.74717796592,
-      "required_pages_per_second_upper_estimate": 0.004030099334955821
+      "required_token_positions_per_second": 9715.026371382244,
+      "required_pages_per_second_upper_estimate": 0.004067478668877674
     },
     {
       "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": 19251.49435593184,
-      "required_pages_per_second_upper_estimate": 0.0040095785782933565
+      "required_token_positions_per_second": 19430.05274276449,
+      "required_pages_per_second_upper_estimate": 0.004046767581369276
     },
     {
       "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": 28877.241533897766,
-      "required_pages_per_second_upper_estimate": 0.001731146536962416
+      "required_token_positions_per_second": 29145.07911414673,
+      "required_pages_per_second_upper_estimate": 0.0017472029909340362
     },
     {
       "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": 28877.241533897766,
-      "required_pages_per_second_upper_estimate": 0.0022450785116986736
+      "required_token_positions_per_second": 29145.07911414673,
+      "required_pages_per_second_upper_estimate": 0.0022659017054700195
     },
     {
       "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": 28877.241533897766,
-      "required_pages_per_second_upper_estimate": 0.37040624843059694
+      "required_token_positions_per_second": 29145.07911414673,
+      "required_pages_per_second_upper_estimate": 0.3738417813284428
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.16557148150809553,
-    "producer_p95_seconds": 0.278401166000549,
-    "gpu_update_seconds": 0.9574321691199345,
-    "producer_batches_per_second": 6.039687456387865,
-    "gpu_batches_per_second": 1.044460414275816,
-    "producer_headroom": 5.782591062175894,
+    "producer_batch_seconds": 0.14302428443767212,
+    "producer_p95_seconds": 0.274942864998593,
+    "gpu_update_seconds": 0.9486335546291221,
+    "producer_batches_per_second": 6.991819633509757,
+    "gpu_batches_per_second": 1.054147826755886,
+    "producer_headroom": 6.632674712262047,
     "sustainable": true,
-    "buffer_seconds": 60.31822665455587,
+    "buffer_seconds": 60.71254749626382,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 246136.36083557722,
-    "upload_bytes_per_second": 4677304.4169802,
+    "generation_bytes_per_second": 273039.4399157559,
+    "upload_bytes_per_second": 4651356.769982685,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.81513327408302,
+    "estimated_full_shards_per_hour": 58.58790777306087,
     "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": 2.320106137998664,
-      "bytes": 11400278,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 1.142467518999183,
       "bytes": 6818785,
@@ -562,6 +556,12 @@
       "bytes": 7546218,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.5842436470011307,
+      "bytes": 5584785,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -301,3 +301,4 @@
 {"step": 915100, "loss": 0.5375716686248779, "lr": 3e-05, "grad_norm": 0.33293834218421364, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133943923327, "tokens_per_second": 144171.80356261818, "data_wait_s": 1.591299951542169e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 915200, "loss": 0.5317983031272888, "lr": 3e-05, "grad_norm": 0.31753135484635203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133958596336, "tokens_per_second": 153445.90836459125, "data_wait_s": 1.0946998372673988e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 915300, "loss": 0.5256280899047852, "lr": 3e-05, "grad_norm": 0.3354857611133877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133973300420, "tokens_per_second": 154145.27959701975, "data_wait_s": 1.274400347028859e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 915400, "loss": 0.49738219380378723, "lr": 3e-05, "grad_norm": 0.3234954893668238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133987942517, "tokens_per_second": 152390.69493372613, "data_wait_s": 1.0581999958958477e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:78b4da74795725b1a0937082e0e794921877ec6f199ec62714e301e2ec46091e
-size 646617
+oid sha256:3219f73b7fe591d940fbefbd1b8d3203de9109ac84d7269e3c259195ebe9970f
+size 648735