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": 76132.59141646765,
-      "required_pages_per_second_upper_estimate": 0.005267029326141207
+      "required_token_positions_per_second": 77032.70419302689,
+      "required_pages_per_second_upper_estimate": 0.005329301216572963
     },
     {
       "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": 9516.573927058456,
-      "required_pages_per_second_upper_estimate": 0.003984390774597589
+      "required_token_positions_per_second": 9629.088024128361,
+      "required_pages_per_second_upper_estimate": 0.004031498077479234
     },
     {
       "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": 19033.147854116913,
-      "required_pages_per_second_upper_estimate": 0.0039641027601497805
+      "required_token_positions_per_second": 19258.176048256722,
+      "required_pages_per_second_upper_estimate": 0.004010970198596553
     },
     {
       "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": 28549.721781175373,
-      "required_pages_per_second_upper_estimate": 0.0017115122278804145
+      "required_token_positions_per_second": 28887.26407238508,
+      "required_pages_per_second_upper_estimate": 0.0017317473728412005
     },
     {
       "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": 28549.721781175373,
-      "required_pages_per_second_upper_estimate": 0.0022196152915316516
+      "required_token_positions_per_second": 28887.26407238508,
+      "required_pages_per_second_upper_estimate": 0.0022458577199815743
     },
     {
       "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": 28549.721781175373,
-      "required_pages_per_second_upper_estimate": 0.36620517670598596
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28887.26407238508,
+      "required_pages_per_second_upper_estimate": 0.3705348067929488
     }
   ],
   "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": 0.12542996475795576,
-    "producer_p95_seconds": 0.22963968999829376,
-    "gpu_update_seconds": 0.9684157419085627,
-    "producer_batches_per_second": 7.9725765843091505,
-    "gpu_batches_per_second": 1.0326143584047804,
-    "producer_headroom": 7.72076866781658,
+    "producer_batch_seconds": 0.1306373438123103,
+    "producer_p95_seconds": 0.23033284699704382,
+    "gpu_update_seconds": 0.9570999846409906,
+    "producer_batches_per_second": 7.654779030387538,
+    "gpu_batches_per_second": 1.0448229192847613,
+    "producer_headroom": 7.326388892414089,
     "sustainable": true,
-    "buffer_seconds": 60.04177599833089,
+    "buffer_seconds": 60.297299032382405,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 265897.4313080301,
-    "upload_bytes_per_second": 4780669.116159606,
+    "generation_bytes_per_second": 298502.71618137974,
+    "upload_bytes_per_second": 4729212.437057524,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.05539898329427,
+    "estimated_full_shards_per_hour": 64.0517341049294,
     "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": 1.4250404260001233,
-      "bytes": 6072497,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.7697906350003905,
       "bytes": 6241903,
@@ -562,6 +556,12 @@
       "bytes": 6312632,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.6241299210014404,
+      "bytes": 2390001,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -205,3 +205,4 @@
 {"step": 945500, "loss": 0.5330240726470947, "lr": 3e-05, "grad_norm": 0.33637224552686323, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138394712044, "tokens_per_second": 153378.28099233445, "data_wait_s": 1.1569001799216494e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 945600, "loss": 0.49929627776145935, "lr": 3e-05, "grad_norm": 0.3394370739139647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138409393627, "tokens_per_second": 153207.73386112833, "data_wait_s": 7.845002983231097e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 945700, "loss": 0.5435240268707275, "lr": 3e-05, "grad_norm": 0.3412607514844836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138424015138, "tokens_per_second": 152529.5826100117, "data_wait_s": 1.0265001037623733e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 945800, "loss": 0.5362918972969055, "lr": 3e-05, "grad_norm": 0.35337433595240636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138438740144, "tokens_per_second": 148556.058747488, "data_wait_s": 1.22519995784387e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:778d0bddcb1248ba111ef8993951864b7a4fda81ece9821658e55e78fd3c9d99
-size 441723
+oid sha256:76bc300fbe102e4960fba6efb40954f93562cc2c0f3ea307e523e907ef7d675c
+size 443841