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": 76791.71823516527,
-      "required_pages_per_second_upper_estimate": 0.005312629248843637
+      "required_token_positions_per_second": 75655.78785758927,
+      "required_pages_per_second_upper_estimate": 0.005234042949601321
     },
     {
       "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": 9598.96477939566,
-      "required_pages_per_second_upper_estimate": 0.004018886103954533
+      "required_token_positions_per_second": 9456.973482198659,
+      "required_pages_per_second_upper_estimate": 0.003959437312933608
     },
     {
       "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": 19197.92955879132,
-      "required_pages_per_second_upper_estimate": 0.003998422443647679
+      "required_token_positions_per_second": 18913.946964397317,
+      "required_pages_per_second_upper_estimate": 0.003939276358360044
     },
     {
       "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": 28796.894338186976,
-      "required_pages_per_second_upper_estimate": 0.0017263298452626183
+      "required_token_positions_per_second": 28370.920446595974,
+      "required_pages_per_second_upper_estimate": 0.0017007933608862095
     },
     {
       "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": 28796.894338186976,
-      "required_pages_per_second_upper_estimate": 0.002238831870642124
+      "required_token_positions_per_second": 28370.920446595974,
+      "required_pages_per_second_upper_estimate": 0.002205714274232058
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28796.894338186976,
-      "required_pages_per_second_upper_estimate": 0.36937564087411623
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28370.920446595974,
+      "required_pages_per_second_upper_estimate": 0.36391170516791693
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.14103641398426703,
-    "producer_p95_seconds": 0.34873290000177803,
-    "gpu_update_seconds": 0.9601035332249891,
-    "producer_batches_per_second": 7.090367457240884,
-    "gpu_batches_per_second": 1.0415543380420638,
-    "producer_headroom": 6.807486847560454,
+    "producer_batch_seconds": 0.10810173653118227,
+    "producer_p95_seconds": 0.3230352110003878,
+    "gpu_update_seconds": 0.9745189639526584,
+    "producer_batches_per_second": 9.250545200183227,
+    "gpu_batches_per_second": 1.026147296245514,
+    "producer_headroom": 9.014831724479796,
     "sustainable": true,
-    "buffer_seconds": 60.486522593174314,
+    "buffer_seconds": 60.42017576506482,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 172,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 546748.263176062,
-    "upload_bytes_per_second": 4795149.072610317,
+    "generation_bytes_per_second": 572781.1879156686,
+    "upload_bytes_per_second": 4848840.847900345,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 117.31944962941547,
+    "estimated_full_shards_per_hour": 61.45275463153145,
     "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": 1.0563156829994114,
-      "bytes": 932363,
-      "pages": 27,
-      "shards": 1
-    },
     {
       "seconds": 1.3606180970018613,
       "bytes": 7382890,
@@ -562,6 +556,12 @@
       "bytes": 1240846,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 1.4493512070002907,
+      "bytes": 7603435,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -329,3 +329,4 @@
 {"step": 957900, "loss": 0.506212055683136, "lr": 3e-05, "grad_norm": 0.3587087317835405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140210585668, "tokens_per_second": 150557.25984268147, "data_wait_s": 1.3520002539735287e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 958000, "loss": 0.5144978761672974, "lr": 3e-05, "grad_norm": 0.3396069652641457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140225274890, "tokens_per_second": 152352.72301761995, "data_wait_s": 1.3647000741912052e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 958100, "loss": 0.5038735866546631, "lr": 3e-05, "grad_norm": 0.33479348778883217, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140239853691, "tokens_per_second": 151053.85292060426, "data_wait_s": 1.4336001186165959e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 958200, "loss": 0.5069932341575623, "lr": 3e-05, "grad_norm": 0.3531510392891549, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140254446158, "tokens_per_second": 152533.25457768576, "data_wait_s": 1.6188001609407365e-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:aa31c2ef50975d802014deb6c7c7ee06db5c3ad6d586a31b592384ae8575a08c
-size 705921
+oid sha256:108887aa16546769af747184c7482e466140b6955de2e53c7eae7451a896e683
+size 708039