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": 76762.8041598951,
-      "required_pages_per_second_upper_estimate": 0.005310628906026548
+      "required_token_positions_per_second": 75529.79605134088,
+      "required_pages_per_second_upper_estimate": 0.0052253265441037896
     },
     {
       "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": 9595.350519986887,
-      "required_pages_per_second_upper_estimate": 0.0040173728890143985
+      "required_token_positions_per_second": 9441.22450641761,
+      "required_pages_per_second_upper_estimate": 0.003952843545650096
     },
     {
       "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": 19190.701039973774,
-      "required_pages_per_second_upper_estimate": 0.003996916933806845
+      "required_token_positions_per_second": 18882.44901283522,
+      "required_pages_per_second_upper_estimate": 0.003932716165706502
     },
     {
       "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": 28786.05155996066,
-      "required_pages_per_second_upper_estimate": 0.0017256798372639144
+      "required_token_positions_per_second": 28323.67351925283,
+      "required_pages_per_second_upper_estimate": 0.0016979609797338688
     },
     {
       "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": 28786.05155996066,
-      "required_pages_per_second_upper_estimate": 0.002237988892327369
+      "required_token_positions_per_second": 28323.67351925283,
+      "required_pages_per_second_upper_estimate": 0.002202041033448399
     },
     {
       "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": 28786.05155996066,
-      "required_pages_per_second_upper_estimate": 0.36923656135709726
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28323.67351925283,
+      "required_pages_per_second_upper_estimate": 0.36330567231375727
     }
   ],
   "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": 0.10171866165632082,
-    "producer_p95_seconds": 0.3487671760012745,
-    "gpu_update_seconds": 0.9604651732944296,
-    "producer_batches_per_second": 9.831037724215474,
-    "gpu_batches_per_second": 1.0411621657971881,
-    "producer_headroom": 9.44236935145269,
+    "producer_batch_seconds": 0.14998238994525082,
+    "producer_p95_seconds": 0.3665946070032078,
+    "gpu_update_seconds": 0.9761445661773518,
+    "producer_batches_per_second": 6.667449427663057,
+    "gpu_batches_per_second": 1.024438423005383,
+    "producer_headroom": 6.508394529075588,
     "sustainable": true,
-    "buffer_seconds": 60.509305917549064,
+    "buffer_seconds": 60.52096310299581,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 310381.71625837556,
-    "upload_bytes_per_second": 5162500.565895141,
+    "generation_bytes_per_second": 353381.1953683351,
+    "upload_bytes_per_second": 5070835.459755014,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.60069099248362,
+    "estimated_full_shards_per_hour": 75.8273782328371,
     "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": 1.5738787880000018,
-      "bytes": 11910990,
-      "pages": 41,
-      "shards": 4
-    },
     {
       "seconds": 1.6920514609992097,
       "bytes": 6058111,
@@ -562,6 +556,12 @@
       "bytes": 10545016,
       "pages": 43,
       "shards": 3
+    },
+    {
+      "seconds": 1.2370521369957714,
+      "bytes": 2360341,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -396,3 +396,4 @@
 {"step": 964600, "loss": 0.530150294303894, "lr": 3e-05, "grad_norm": 0.34140706599839477, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141190980952, "tokens_per_second": 149598.39953637886, "data_wait_s": 1.2651995348278433e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 964700, "loss": 0.5566827654838562, "lr": 3e-05, "grad_norm": 0.35749923745884793, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141205627971, "tokens_per_second": 152080.44458582226, "data_wait_s": 1.727099879644811e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 964800, "loss": 0.5158869028091431, "lr": 3e-05, "grad_norm": 0.33215481712055117, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141220304902, "tokens_per_second": 152368.5144324505, "data_wait_s": 1.2736003554891795e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 964900, "loss": 0.5310418605804443, "lr": 3e-05, "grad_norm": 0.3544124334635774, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141234854891, "tokens_per_second": 151755.32677238516, "data_wait_s": 9.47499938774854e-06, "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:d9bba3e9a5cb34cd4c7d9526baba89f50b90a89de89a8836215e84037599b8c6
-size 848610
+oid sha256:7b8edbd88d10c6084c81c1ffff03a5162359d376015503cc2a552b9572102416
+size 850728