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": 74952.07022406434,
-      "required_pages_per_second_upper_estimate": 0.006496655482736652
+      "required_token_positions_per_second": 75275.1693256176,
+      "required_pages_per_second_upper_estimate": 0.006524660893972102
     },
     {
       "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": 9369.008778008043,
-      "required_pages_per_second_upper_estimate": 0.0037374376807116816
+      "required_token_positions_per_second": 9409.3961657022,
+      "required_pages_per_second_upper_estimate": 0.0037535488135081374
     },
     {
       "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": 18738.017556016086,
-      "required_pages_per_second_upper_estimate": 0.004334710583738648
+      "required_token_positions_per_second": 18818.7923314044,
+      "required_pages_per_second_upper_estimate": 0.0043533964064372405
     },
     {
       "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": 28107.026334024125,
-      "required_pages_per_second_upper_estimate": 0.0017115704375716809
+      "required_token_positions_per_second": 28228.188497106596,
+      "required_pages_per_second_upper_estimate": 0.0017189485776146621
     },
     {
       "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": 28107.026334024125,
-      "required_pages_per_second_upper_estimate": 0.002163756692421826
+      "required_token_positions_per_second": 28228.188497106596,
+      "required_pages_per_second_upper_estimate": 0.002173084091134248
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28107.026334024125,
-      "required_pages_per_second_upper_estimate": 0.3489951989026674
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28228.188497106596,
+      "required_pages_per_second_upper_estimate": 0.35049962746447716
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12377871077342206,
-    "producer_p95_seconds": 0.31106313400232466,
-    "gpu_update_seconds": 0.9836686268917582,
-    "producer_batches_per_second": 8.078933717693248,
-    "gpu_batches_per_second": 1.016602514974831,
-    "producer_headroom": 7.946993636832844,
+    "producer_batch_seconds": 0.09941502975789263,
+    "producer_p95_seconds": 0.28495891600323375,
+    "gpu_update_seconds": 0.9794464849501034,
+    "producer_batches_per_second": 10.05884122788395,
+    "gpu_batches_per_second": 1.0209848270076172,
+    "producer_headroom": 9.852096683322117,
     "sustainable": true,
-    "buffer_seconds": 60.00378624039725,
+    "buffer_seconds": 60.725682066906415,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 204745.3781503934,
-    "upload_bytes_per_second": 6168580.783124281,
+    "generation_bytes_per_second": 204627.30131702265,
+    "upload_bytes_per_second": 6236807.470095545,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 43.933591922606006,
+    "estimated_full_shards_per_hour": 43.908255382852644,
     "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": 1.028129626000009,
-      "bytes": 6087526,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.7597279170004185,
       "bytes": 2156570,
@@ -562,6 +556,12 @@
       "bytes": 2201439,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.2722073730001284,
+      "bytes": 12361137,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -160,3 +160,4 @@
 {"step": 16000, "loss": 0.7325483039021492, "lr": 0.0003, "grad_norm": 0.20352032907254936, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2341029817, "tokens_per_second": 145638.20763586988, "data_wait_s": 1.105499904952012e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 16100, "loss": 0.6394673865288496, "lr": 0.0003, "grad_norm": 0.13619027882804471, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2355617913, "tokens_per_second": 149055.33354085588, "data_wait_s": 9.326999133918434e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 16200, "loss": 0.7200316488742828, "lr": 0.0003, "grad_norm": 0.14547465273479035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2370236048, "tokens_per_second": 147838.81526074515, "data_wait_s": 1.0864998330362141e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 16300, "loss": 0.6719108186662197, "lr": 0.0003, "grad_norm": 0.13416477956147393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2384862018, "tokens_per_second": 148957.66477971146, "data_wait_s": 1.0330997611163184e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c0e2f0e142490ae4680e62d16dd05404bceb521bdf1108395f51756f656e2a83
-size 339891
+oid sha256:1df53ccf96db6239e76e05d98bcba773455d067feaa9de564b7131bac4ebfef6
+size 341974