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": 73907.42228458513,
-      "required_pages_per_second_upper_estimate": 0.006406108046978591
+      "required_token_positions_per_second": 75143.27492024105,
+      "required_pages_per_second_upper_estimate": 0.006513228621197377
     },
     {
       "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": 9238.427785573142,
-      "required_pages_per_second_upper_estimate": 0.003685346970469579
+      "required_token_positions_per_second": 9392.909365030131,
+      "required_pages_per_second_upper_estimate": 0.003746971982220413
     },
     {
       "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": 18476.855571146283,
-      "required_pages_per_second_upper_estimate": 0.0042742953548329694
+      "required_token_positions_per_second": 18785.818730060262,
+      "required_pages_per_second_upper_estimate": 0.004345768544081838
     },
     {
       "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": 27715.283356719425,
-      "required_pages_per_second_upper_estimate": 0.001687715345570383
+      "required_token_positions_per_second": 28178.728095090395,
+      "required_pages_per_second_upper_estimate": 0.0017159366986305498
     },
     {
       "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": 27715.283356719425,
-      "required_pages_per_second_upper_estimate": 0.0021335992336149433
+      "required_token_positions_per_second": 28178.728095090395,
+      "required_pages_per_second_upper_estimate": 0.0021692764924718857
     },
     {
       "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": 27715.283356719425,
-      "required_pages_per_second_upper_estimate": 0.3441310622039488
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28178.728095090395,
+      "required_pages_per_second_upper_estimate": 0.34988549480475306
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.124362981648801,
-    "producer_p95_seconds": 0.40652090799994767,
-    "gpu_update_seconds": 0.997572337404838,
-    "producer_batches_per_second": 8.040978004403138,
-    "gpu_batches_per_second": 1.0024335704831968,
-    "producer_headroom": 8.021457222873329,
+    "producer_batch_seconds": 0.1239068050865626,
+    "producer_p95_seconds": 0.40487155500159133,
+    "gpu_update_seconds": 0.9811656476012889,
+    "producer_batches_per_second": 8.070581751352474,
+    "gpu_batches_per_second": 1.0191958946430264,
+    "producer_headroom": 7.918577570584894,
     "sustainable": true,
-    "buffer_seconds": 60.85191258169512,
+    "buffer_seconds": 60.832270151279914,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 320619.61218343594,
+    "generation_bytes_per_second": 321861.3297646052,
     "upload_bytes_per_second": 6292706.395301407,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.79750513198194,
+    "estimated_full_shards_per_hour": 69.06394881919496,
     "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-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -380,3 +380,4 @@
 {"step": 78000, "loss": 0.6717341057956219, "lr": 0.0003, "grad_norm": 0.1393901915487311, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11415141762, "tokens_per_second": 149919.69667152414, "data_wait_s": 1.1878997611347586e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 78100, "loss": 0.6047646552324295, "lr": 0.0003, "grad_norm": 0.13520182350403326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11429812036, "tokens_per_second": 148336.8116493828, "data_wait_s": 1.0003001079894602e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 78200, "loss": 0.6694426946341991, "lr": 0.0003, "grad_norm": 0.14425529049851427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11444504104, "tokens_per_second": 150539.53375494335, "data_wait_s": 1.3578996004071087e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 78300, "loss": 0.5956331938505173, "lr": 0.0003, "grad_norm": 0.12783067811872012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11459096685, "tokens_per_second": 148629.83206661104, "data_wait_s": 1.3771001249551773e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:937bae56d83bd8818acdab93384e7c4c2aea79f1a6e43ec881cf838e39cb8b80
-size 814722
+oid sha256:bfa05db6fc58c67e7d0266cf653dedd80f7fc8fd77b2c4e8009b6d7a898ae839
+size 816840