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": 75550.89638149532,
-      "required_pages_per_second_upper_estimate": 0.006548560216351748
+      "required_token_positions_per_second": 74135.9714824822,
+      "required_pages_per_second_upper_estimate": 0.006425918112199663
     },
     {
       "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": 9443.862047686915,
-      "required_pages_per_second_upper_estimate": 0.0037672977691426976
+      "required_token_positions_per_second": 9266.996435310275,
+      "required_pages_per_second_upper_estimate": 0.0036967434319890996
     },
     {
       "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": 18887.72409537383,
-      "required_pages_per_second_upper_estimate": 0.004369342556873957
+      "required_token_positions_per_second": 18533.99287062055,
+      "required_pages_per_second_upper_estimate": 0.004287513063484215
     },
     {
       "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": 28331.58614306074,
-      "required_pages_per_second_upper_estimate": 0.0017252449517677468
+      "required_token_positions_per_second": 27800.989305930823,
+      "required_pages_per_second_upper_estimate": 0.0016929343882129938
     },
     {
       "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": 28331.58614306074,
-      "required_pages_per_second_upper_estimate": 0.0021810439281428015
+      "required_token_positions_per_second": 27800.989305930823,
+      "required_pages_per_second_upper_estimate": 0.0021401971148344886
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28331.58614306074,
-      "required_pages_per_second_upper_estimate": 0.35178348017756733
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27800.989305930823,
+      "required_pages_per_second_upper_estimate": 0.3451952432537942
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09844005843743275,
-    "producer_p95_seconds": 0.3069954200000211,
-    "gpu_update_seconds": 0.9758719423752358,
-    "producer_batches_per_second": 10.158466135364876,
-    "gpu_batches_per_second": 1.0247246145493614,
-    "producer_headroom": 9.913362079071575,
+    "producer_batch_seconds": 0.4993493134295477,
+    "producer_p95_seconds": 0.32166762800261495,
+    "gpu_update_seconds": 0.9944969833897894,
+    "producer_batches_per_second": 2.002606137839595,
+    "gpu_batches_per_second": 1.0055334673730767,
+    "producer_headroom": 1.9915857629993543,
     "sustainable": true,
-    "buffer_seconds": 60.50406042726462,
+    "buffer_seconds": 60.664315986777154,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 528612.4619241259,
-    "upload_bytes_per_second": 3608489.7042848063,
+    "generation_bytes_per_second": 474233.45898903796,
+    "upload_bytes_per_second": 3573120.1789109656,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.42792886059601,
+    "estimated_full_shards_per_hour": 101.75946070912698,
     "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": 1.222915964999629,
-      "bytes": 5976864,
-      "pages": 26,
-      "shards": 2
-    },
     {
       "seconds": 1.3758100840004772,
       "bytes": 11655433,
@@ -562,6 +556,12 @@
       "bytes": 5786171,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.022158355001011,
+      "bytes": 1083307,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -232,3 +232,4 @@
 {"step": 218200, "loss": 0.5977561417967081, "lr": 0.0003, "grad_norm": 0.15101460923973592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31937438736, "tokens_per_second": 147734.543121157, "data_wait_s": 2.2462998458649963e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 218300, "loss": 0.5979484468698502, "lr": 0.0003, "grad_norm": 0.14897629113257588, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31952163229, "tokens_per_second": 149968.19132002458, "data_wait_s": 1.2033000530209392e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 218400, "loss": 0.5609187074005604, "lr": 0.0003, "grad_norm": 0.15159725522328146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31966915344, "tokens_per_second": 149543.51853069908, "data_wait_s": 1.281100048799999e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 218500, "loss": 0.5691029401496053, "lr": 0.0003, "grad_norm": 0.13410400851578652, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31981575546, "tokens_per_second": 148790.43486262104, "data_wait_s": 1.061799775925465e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6c86e73697993f64255f3504de71e142b34abe311866920ea2b09c58f360a8d2
-size 498909
+oid sha256:d7d3794ec4fb4b8622f1dd9a74450b1c6627894d79b0ba15d154478548bca9d6
+size 501027