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": 76242.52092923179,
-      "required_pages_per_second_upper_estimate": 0.0066085084792431555
+      "required_token_positions_per_second": 76112.91550812153,
+      "required_pages_per_second_upper_estimate": 0.006597274609823255
     },
     {
       "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": 9530.315116153974,
-      "required_pages_per_second_upper_estimate": 0.003801785190742769
+      "required_token_positions_per_second": 9514.114438515191,
+      "required_pages_per_second_upper_estimate": 0.003795322498210943
     },
     {
       "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": 19060.630232307947,
-      "required_pages_per_second_upper_estimate": 0.004409341348609587
+      "required_token_positions_per_second": 19028.228877030382,
+      "required_pages_per_second_upper_estimate": 0.004401845865310499
     },
     {
       "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": 28590.945348461923,
-      "required_pages_per_second_upper_estimate": 0.0017410385666240924
+      "required_token_positions_per_second": 28542.34331554557,
+      "required_pages_per_second_upper_estimate": 0.0017380789578146463
     },
     {
       "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": 28590.945348461923,
-      "required_pages_per_second_upper_estimate": 0.002201010117726816
+      "required_token_positions_per_second": 28542.34331554557,
+      "required_pages_per_second_upper_estimate": 0.0021972685986938775
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28590.945348461923,
-      "required_pages_per_second_upper_estimate": 0.3550038534892276
+      "required_token_positions_per_second": 28542.34331554557,
+      "required_pages_per_second_upper_estimate": 0.35440037890094706
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1327493430001141,
-    "producer_p95_seconds": 0.3322277670013136,
-    "gpu_update_seconds": 0.967019441401134,
-    "producer_batches_per_second": 7.53299396742883,
-    "gpu_batches_per_second": 1.0341053728465683,
-    "producer_headroom": 7.28455161846114,
+    "producer_batch_seconds": 0.10867654062485599,
+    "producer_p95_seconds": 0.33580439099750947,
+    "gpu_update_seconds": 0.9686660865347216,
+    "producer_batches_per_second": 9.201617885979015,
+    "gpu_batches_per_second": 1.0323474868180544,
+    "producer_headroom": 8.913295187399191,
     "sustainable": true,
-    "buffer_seconds": 60.922224808271444,
+    "buffer_seconds": 60.05729736515274,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 395252.12788003165,
+    "generation_bytes_per_second": 388221.2361143507,
     "upload_bytes_per_second": 4259802.362978847,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.8119056444236,
+    "estimated_full_shards_per_hour": 83.30323994229212,
     "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": [

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -344,3 +344,4 @@
 {"step": 189400, "loss": 0.5764011181890965, "lr": 0.0003, "grad_norm": 0.1454088546957294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27719889100, "tokens_per_second": 151355.7005699601, "data_wait_s": 1.0283998562954366e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 189500, "loss": 0.5868939906358719, "lr": 0.0003, "grad_norm": 0.1563980391468381, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27734569853, "tokens_per_second": 151238.77286017945, "data_wait_s": 1.2299999070819467e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 189600, "loss": 0.6668337341398001, "lr": 0.0003, "grad_norm": 0.2089877609521357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27749200103, "tokens_per_second": 149880.15524852555, "data_wait_s": 1.073500607162714e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 189700, "loss": 0.5790050812065601, "lr": 0.0003, "grad_norm": 0.14124361286865064, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27764026242, "tokens_per_second": 152038.68240380072, "data_wait_s": 9.209004929289222e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e753e76a9bea40d213b280690e457b569f5c403f9807734e5624c3e4ae49c715
-size 737691
+oid sha256:f5c0f81d3accc19e0e87c77fdcffe5fa799cd8ef57ab9965ccbe5476a82bd04a
+size 739809