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": 75700.32285632046,
-      "required_pages_per_second_upper_estimate": 0.006561512124471654
+      "required_token_positions_per_second": 73958.18411879856,
+      "required_pages_per_second_upper_estimate": 0.0064105079541135175
     },
     {
       "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": 9462.540357040058,
-      "required_pages_per_second_upper_estimate": 0.003774748826009278
+      "required_token_positions_per_second": 9244.77301484982,
+      "required_pages_per_second_upper_estimate": 0.003687878177297678
     },
     {
       "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": 18925.080714080115,
-      "required_pages_per_second_upper_estimate": 0.004377984353157622
+      "required_token_positions_per_second": 18489.54602969964,
+      "required_pages_per_second_upper_estimate": 0.004277231069074846
     },
     {
       "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": 28387.62107112017,
-      "required_pages_per_second_upper_estimate": 0.0017286571848940193
+      "required_token_positions_per_second": 27734.31904454946,
+      "required_pages_per_second_upper_estimate": 0.0016888745190866967
     },
     {
       "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": 28387.62107112017,
-      "required_pages_per_second_upper_estimate": 0.002185357652019425
+      "required_token_positions_per_second": 27734.31904454946,
+      "required_pages_per_second_upper_estimate": 0.002135064653558256
     },
     {
       "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": 28387.62107112017,
-      "required_pages_per_second_upper_estimate": 0.3524792464472251
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27734.31904454946,
+      "required_pages_per_second_upper_estimate": 0.34436742173844886
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.2644437009218734,
-    "producer_p95_seconds": 0.16535117300009006,
-    "gpu_update_seconds": 0.9739456480249902,
-    "producer_batches_per_second": 3.781523237323915,
-    "gpu_batches_per_second": 1.026751340824659,
-    "producer_headroom": 3.6829980998969996,
+    "producer_batch_seconds": 0.07882897369533026,
+    "producer_p95_seconds": 0.16976602200020352,
+    "gpu_update_seconds": 0.9968876450721287,
+    "producer_batches_per_second": 12.685690972775392,
+    "gpu_batches_per_second": 1.0031220719238085,
+    "producer_headroom": 12.646208599962822,
     "sustainable": true,
-    "buffer_seconds": 60.38463017754939,
+    "buffer_seconds": 60.81014634939985,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 516494.2735893789,
-    "upload_bytes_per_second": 4500469.795094228,
+    "generation_bytes_per_second": 515543.20450798137,
+    "upload_bytes_per_second": 4394505.093192198,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.82764774094606,
+    "estimated_full_shards_per_hour": 110.62357045583325,
     "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": [
@@ -322,6 +322,12 @@
       "bytes": 10860861,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 0.9114932880002016,
+      "bytes": 506194,
+      "pages": 16,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -27,3 +27,4 @@
 {"step": 239800, "loss": 0.6235589403659105, "lr": 0.0003, "grad_norm": 0.15547576243761393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35099075885, "tokens_per_second": 146513.72239722084, "data_wait_s": 1.4210999779606936e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 239900, "loss": 0.6314800642430782, "lr": 0.0003, "grad_norm": 0.16600692105706924, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35113637319, "tokens_per_second": 148075.33592057668, "data_wait_s": 1.1266000001342036e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 240000, "loss": 0.5885987225919962, "lr": 0.0003, "grad_norm": 0.17048077004762824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35128325853, "tokens_per_second": 147986.62644008137, "data_wait_s": 1.0194999958912376e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 240100, "loss": 0.5598774999380112, "lr": 0.0003, "grad_norm": 0.15279531327039617, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35143001793, "tokens_per_second": 130649.21440094686, "data_wait_s": 9.452000085730106e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4bf0bed68778a5040297f6de599c56cff3d4333e7ed02d06ad78a1dceedce948
-size 62370
+oid sha256:97fdcfc07073407cc1176f6a93af153f06e9003046bd350c1aaca952e3fda6f7
+size 64488