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": 75458.95402589241,
-      "required_pages_per_second_upper_estimate": 0.006540590885993859
+      "required_token_positions_per_second": 74222.010950236,
+      "required_pages_per_second_upper_estimate": 0.006433375795199524
     },
     {
       "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": 9432.369253236551,
-      "required_pages_per_second_upper_estimate": 0.003762713121603858
+      "required_token_positions_per_second": 9277.7513687795,
+      "required_pages_per_second_upper_estimate": 0.0037010337357505584
     },
     {
       "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": 18864.738506473102,
-      "required_pages_per_second_upper_estimate": 0.004364025245414321
+      "required_token_positions_per_second": 18555.502737559,
+      "required_pages_per_second_upper_estimate": 0.004292488992639697
     },
     {
       "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": 28297.107759709652,
-      "required_pages_per_second_upper_estimate": 0.0017231453991157627
+      "required_token_positions_per_second": 27833.2541063385,
+      "required_pages_per_second_upper_estimate": 0.0016948991452774961
     },
     {
       "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": 28297.107759709652,
-      "required_pages_per_second_upper_estimate": 0.002178389686750161
+      "required_token_positions_per_second": 27833.2541063385,
+      "required_pages_per_second_upper_estimate": 0.0021426809484845558
     },
     {
       "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": 28297.107759709652,
-      "required_pages_per_second_upper_estimate": 0.35135537404807293
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27833.2541063385,
+      "required_pages_per_second_upper_estimate": 0.34559586409151694
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.12865515918730352,
-    "producer_p95_seconds": 0.34807392400034587,
-    "gpu_update_seconds": 0.9770609856943092,
-    "producer_batches_per_second": 7.772715888868031,
-    "gpu_batches_per_second": 1.0234775665404243,
-    "producer_headroom": 7.594417447899217,
+    "producer_batch_seconds": 0.20855013382040966,
+    "producer_p95_seconds": 0.35660255100083305,
+    "gpu_update_seconds": 0.993344144898375,
+    "producer_batches_per_second": 4.79501010946911,
+    "gpu_batches_per_second": 1.0067004523415255,
+    "producer_headroom": 4.763095216969656,
     "sustainable": true,
-    "buffer_seconds": 60.57778111304717,
+    "buffer_seconds": 60.59399283880087,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 371741.274192532,
-    "upload_bytes_per_second": 4161727.9075428573,
+    "generation_bytes_per_second": 374795.5954121424,
+    "upload_bytes_per_second": 4189189.617109923,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.76702374774904,
+    "estimated_full_shards_per_hour": 80.42240998051838,
     "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": 0.9606277580023743,
-      "bytes": 2676212,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.570391033001215,
       "bytes": 9821415,
@@ -562,6 +556,12 @@
       "bytes": 7895099,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.032417123002233,
+      "bytes": 5759947,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -308,3 +308,4 @@
 {"step": 345800, "loss": 0.5773425437510014, "lr": 0.0003, "grad_norm": 0.1719673929543946, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50615996588, "tokens_per_second": 149396.85238823932, "data_wait_s": 1.3353997928788885e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 345900, "loss": 0.6154396384954453, "lr": 0.0003, "grad_norm": 0.15911986876923162, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50630570454, "tokens_per_second": 148408.74728900185, "data_wait_s": 1.6066001990111545e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 346000, "loss": 0.6023015528917313, "lr": 0.0003, "grad_norm": 0.1422033520210749, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50645356753, "tokens_per_second": 150134.3683929201, "data_wait_s": 9.304996638093144e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 346100, "loss": 0.607675364241004, "lr": 0.0003, "grad_norm": 0.16171144785741007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50660075345, "tokens_per_second": 149965.76371127076, "data_wait_s": 1.42700009746477e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:136b33ef4370c6dab06c4132a254f404736a1b31b436382b34de7a93355baf31
-size 661443
+oid sha256:965e7be98b51d516cc5a530c241166290ed660ab4a2f22023b4c97ef1bab4743
+size 663561