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": 76801.10612007229,
-      "required_pages_per_second_upper_estimate": 0.005313278724504398
+      "required_token_positions_per_second": 77459.17729436498,
+      "required_pages_per_second_upper_estimate": 0.005358805615277461
     },
     {
       "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": 9600.138265009036,
-      "required_pages_per_second_upper_estimate": 0.004019377417875687
+      "required_token_positions_per_second": 9682.397161795623,
+      "required_pages_per_second_upper_estimate": 0.004053817500199133
     },
     {
       "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": 19200.276530018073,
-      "required_pages_per_second_upper_estimate": 0.003998911255860419
+      "required_token_positions_per_second": 19364.794323591246,
+      "required_pages_per_second_upper_estimate": 0.004033175973635734
     },
     {
       "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": 28800.414795027107,
-      "required_pages_per_second_upper_estimate": 0.001726540891274758
+      "required_token_positions_per_second": 29047.19148538687,
+      "required_pages_per_second_upper_estimate": 0.0017413347770556373
     },
     {
       "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": 28800.414795027107,
-      "required_pages_per_second_upper_estimate": 0.002239105570676591
+      "required_token_positions_per_second": 29047.19148538687,
+      "required_pages_per_second_upper_estimate": 0.0022582913728961163
     },
     {
       "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": 28800.414795027107,
-      "required_pages_per_second_upper_estimate": 0.36942079751448936
+      "required_token_positions_per_second": 29047.19148538687,
+      "required_pages_per_second_upper_estimate": 0.37258618393025833
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.050190285085960795,
-    "producer_p95_seconds": 0.1691902399998071,
-    "gpu_update_seconds": 0.9599861736982311,
-    "producer_batches_per_second": 19.92417453471926,
-    "gpu_batches_per_second": 1.0416816693803208,
-    "producer_headroom": 19.126932075680877,
+    "producer_batch_seconds": 0.06658308002339908,
+    "producer_p95_seconds": 0.16902427300010459,
+    "gpu_update_seconds": 0.9518304037727442,
+    "producer_batches_per_second": 15.018830604540572,
+    "gpu_batches_per_second": 1.0506073309240043,
+    "producer_headroom": 14.2953795985143,
     "sustainable": true,
-    "buffer_seconds": 60.47912894298856,
+    "buffer_seconds": 60.91714584145563,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 499106.8060925204,
-    "upload_bytes_per_second": 2028727.9608457147,
+    "generation_bytes_per_second": 520660.8256385236,
+    "upload_bytes_per_second": 2129636.277989188,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.09670197564802,
+    "estimated_full_shards_per_hour": 111.72169281832487,
     "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": [
@@ -256,6 +256,12 @@
       "bytes": 7577595,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.9591294869997,
+      "bytes": 6044241,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -56,3 +56,4 @@
 {"step": 690600, "loss": 0.5559753179550171, "lr": 8e-05, "grad_norm": 0.18827603670183796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101085098989, "tokens_per_second": 153872.65503392747, "data_wait_s": 1.7520999790576752e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 690700, "loss": 0.5771751403808594, "lr": 8e-05, "grad_norm": 0.20993416394020747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101099755045, "tokens_per_second": 152782.80627288757, "data_wait_s": 1.4968999494158197e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 690800, "loss": 0.5475606918334961, "lr": 8e-05, "grad_norm": 0.20118566788462452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101114401085, "tokens_per_second": 153311.38541503355, "data_wait_s": 1.3629000022774562e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 690900, "loss": 0.5649669170379639, "lr": 8e-05, "grad_norm": 0.20676005263549074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101129081222, "tokens_per_second": 153118.6356530891, "data_wait_s": 1.3066999599686824e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:45bffe329ccaa0067d6c0db7f93cefe8556f3a39c5b0ca8db9677106ae4512f4
-size 123792
+oid sha256:d22cb36502469e1703cfd7b71210a530ee7f4e427309cb8f3e617e78dd5152b7
+size 125910