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": 76450.62966049743,
-      "required_pages_per_second_upper_estimate": 0.005289031949813583
+      "required_token_positions_per_second": 77579.11526486342,
+      "required_pages_per_second_upper_estimate": 0.005367103202370976
     },
     {
       "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": 9556.32870756218,
-      "required_pages_per_second_upper_estimate": 0.004001035271020279
+      "required_token_positions_per_second": 9697.389408107927,
+      "required_pages_per_second_upper_estimate": 0.004060094440656391
     },
     {
       "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": 19112.65741512436,
-      "required_pages_per_second_upper_estimate": 0.0039806625048995035
+      "required_token_positions_per_second": 19394.778816215854,
+      "required_pages_per_second_upper_estimate": 0.004039420952705194
     },
     {
       "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": 28668.98612268654,
-      "required_pages_per_second_upper_estimate": 0.0017186619430479039
+      "required_token_positions_per_second": 29092.16822432378,
+      "required_pages_per_second_upper_estimate": 0.0017440310638794012
     },
     {
       "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": 28668.98612268654,
-      "required_pages_per_second_upper_estimate": 0.0022288875694957467
+      "required_token_positions_per_second": 29092.16822432378,
+      "required_pages_per_second_upper_estimate": 0.002261788116516694
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28668.98612268654,
-      "required_pages_per_second_upper_estimate": 0.3677349716228183
+      "required_token_positions_per_second": 29092.16822432378,
+      "required_pages_per_second_upper_estimate": 0.3731630972450813
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08406115645304624,
-    "producer_p95_seconds": 0.2444274330009648,
-    "gpu_update_seconds": 0.9643870865081411,
-    "producer_batches_per_second": 11.896100912655974,
-    "gpu_batches_per_second": 1.036928028164299,
-    "producer_headroom": 11.472446099963133,
+    "producer_batch_seconds": 0.0799293461718662,
+    "producer_p95_seconds": 0.22560693000013998,
+    "gpu_update_seconds": 0.9503588658917378,
+    "producer_batches_per_second": 12.511049419192965,
+    "gpu_batches_per_second": 1.0522340937617107,
+    "producer_headroom": 11.88998673713971,
     "sustainable": true,
-    "buffer_seconds": 60.75638645001289,
+    "buffer_seconds": 60.82296741707122,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 333504.99090367305,
-    "upload_bytes_per_second": 5144267.201871949,
+    "generation_bytes_per_second": 319801.76732150296,
+    "upload_bytes_per_second": 5185689.43209054,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.56240744907993,
+    "estimated_full_shards_per_hour": 68.62201466306512,
     "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": 0.9512228559997311,
-      "bytes": 901267,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.6983987419989717,
       "bytes": 12559131,
@@ -562,6 +556,12 @@
       "bytes": 2462043,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.2794374899985996,
+      "bytes": 6045695,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -162,3 +162,4 @@
 {"step": 901200, "loss": 0.5257138013839722, "lr": 3e-05, "grad_norm": 0.3073814167022591, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131910073218, "tokens_per_second": 152015.55721943942, "data_wait_s": 1.7049000234692357e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 901300, "loss": 0.5314346551895142, "lr": 3e-05, "grad_norm": 0.34171601287333275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131924822827, "tokens_per_second": 153930.11609007986, "data_wait_s": 1.2012998922728002e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 901400, "loss": 0.5353484749794006, "lr": 3e-05, "grad_norm": 0.3146727884420038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131939470110, "tokens_per_second": 153312.7049090487, "data_wait_s": 9.49600143940188e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 901500, "loss": 0.5521841049194336, "lr": 3e-05, "grad_norm": 0.3143202209296977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131953978946, "tokens_per_second": 150168.55357323462, "data_wait_s": 1.8426999304210767e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:42d6060ded010e0a1734d5ee6b9afd331d5cb68183dd86bffb47552a34c0a3e6
-size 349866
+oid sha256:e2267eefdb80361884d9ef4efc268427a5441284fd852a88ded0275db22ffcef
+size 351984