Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76068.57268524778,
-      "required_pages_per_second_upper_estimate": 0.005262600361771495
+      "required_token_positions_per_second": 76528.43666681554,
+      "required_pages_per_second_upper_estimate": 0.005294414819047798
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9508.571585655973,
-      "required_pages_per_second_upper_estimate": 0.00398104036136026
+      "required_token_positions_per_second": 9566.054583351943,
+      "required_pages_per_second_upper_estimate": 0.004005107292114071
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19017.143171311945,
-      "required_pages_per_second_upper_estimate": 0.003960769406793374
+      "required_token_positions_per_second": 19132.109166703885,
+      "required_pages_per_second_upper_estimate": 0.003984713791776334
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28525.71475696792,
-      "required_pages_per_second_upper_estimate": 0.0017100730434357811
+      "required_token_positions_per_second": 28698.163750055824,
+      "required_pages_per_second_upper_estimate": 0.0017204110972570351
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28525.71475696792,
-      "required_pages_per_second_upper_estimate": 0.0022177488509952615
+      "required_token_positions_per_second": 28698.163750055824,
+      "required_pages_per_second_upper_estimate": 0.002231156001684889
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28525.71475696792,
-      "required_pages_per_second_upper_estimate": 0.36589724037618704
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28698.163750055824,
+      "required_pages_per_second_upper_estimate": 0.36810923089821607
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2073915345547448,
-    "producer_p95_seconds": 0.3122991260024719,
-    "gpu_update_seconds": 0.969230753218777,
-    "producer_batches_per_second": 4.821797582755393,
-    "gpu_batches_per_second": 1.0317460487907957,
-    "producer_headroom": 4.673434503002488,
+    "producer_batch_seconds": 0.1191363028205501,
+    "producer_p95_seconds": 0.3289518569945358,
+    "gpu_update_seconds": 0.9634065872924088,
+    "producer_batches_per_second": 8.393747131017294,
+    "gpu_batches_per_second": 1.0379833532282923,
+    "producer_headroom": 8.08659127808882,
     "sustainable": true,
-    "buffer_seconds": 60.092306699564176,
+    "buffer_seconds": 60.69461499942175,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 168,
+    "upload_pages_per_commit": 164,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 654255.9418886142,
-    "upload_bytes_per_second": 5430522.109190463,
+    "generation_bytes_per_second": 645988.2628555205,
+    "upload_bytes_per_second": 5431514.106085738,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.19404741522703,
+    "estimated_full_shards_per_hour": 69.30702168583494,
     "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": 1.1124823669997568,
-      "bytes": 8182028,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.1143527310014179,
       "bytes": 6164100,
@@ -562,6 +556,12 @@
       "bytes": 6046016,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.4766436999998405,
+      "bytes": 10238885,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -354,3 +354,4 @@
 {"step": 680400, "loss": 0.6108964681625366, "lr": 0.0001, "grad_norm": 0.24687034693302684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99592982197, "tokens_per_second": 152342.38607546993, "data_wait_s": 9.714000043459237e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 680500, "loss": 0.592078447341919, "lr": 0.0001, "grad_norm": 0.2525708030391718, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99607513653, "tokens_per_second": 150977.59807706944, "data_wait_s": 1.757699646987021e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 680600, "loss": 0.5751513838768005, "lr": 0.0001, "grad_norm": 0.20857990913504343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99622064215, "tokens_per_second": 151103.80003626796, "data_wait_s": 1.5407000319100916e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 680700, "loss": 0.5364224910736084, "lr": 0.0001, "grad_norm": 0.2100102845041653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99636574961, "tokens_per_second": 151614.97339502323, "data_wait_s": 1.017199974739924e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:897d1b27718e7f1fcbf496780a7d01d5aced805083ed6b57174e8352b32d94be
-size 759654
+oid sha256:defcd045026d092c94f635aa8f42d57a3c08bf0138c0ccf6bb4f96aef1bc5604
+size 761772