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": 76589.5404033501,
-      "required_pages_per_second_upper_estimate": 0.005298642117321461
+      "required_token_positions_per_second": 75902.32818565596,
+      "required_pages_per_second_upper_estimate": 0.00525109918154935
     },
     {
       "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": 9573.692550418762,
-      "required_pages_per_second_upper_estimate": 0.0040083051494260585
+      "required_token_positions_per_second": 9487.791023206995,
+      "required_pages_per_second_upper_estimate": 0.003972339973810363
     },
     {
       "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": 19147.385100837524,
-      "required_pages_per_second_upper_estimate": 0.003987895366002896
+      "required_token_positions_per_second": 18975.58204641399,
+      "required_pages_per_second_upper_estimate": 0.003952113320517699
     },
     {
       "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": 28721.077651256288,
-      "required_pages_per_second_upper_estimate": 0.0017217847506465016
+      "required_token_positions_per_second": 28463.37306962099,
+      "required_pages_per_second_upper_estimate": 0.0017063357544695778
     },
     {
       "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": 28721.077651256288,
-      "required_pages_per_second_upper_estimate": 0.0022329374567156253
+      "required_token_positions_per_second": 28463.37306962099,
+      "required_pages_per_second_upper_estimate": 0.002212902059016141
     },
     {
       "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": 28721.077651256288,
-      "required_pages_per_second_upper_estimate": 0.3684031458197854
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28463.37306962099,
+      "required_pages_per_second_upper_estimate": 0.3650975881481893
     }
   ],
   "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": 0.12844523256222828,
-    "producer_p95_seconds": 0.2855826149971108,
-    "gpu_update_seconds": 0.9626379739546663,
-    "producer_batches_per_second": 7.785419357744763,
-    "gpu_batches_per_second": 1.0388121256964804,
-    "producer_headroom": 7.494540316926858,
+    "producer_batch_seconds": 0.5438251753670897,
+    "producer_p95_seconds": 0.26902029600023525,
+    "gpu_update_seconds": 0.9713536035371986,
+    "producer_batches_per_second": 1.8388262355176657,
+    "gpu_batches_per_second": 1.0294912134556202,
+    "producer_headroom": 1.7861504901488259,
     "sustainable": true,
-    "buffer_seconds": 60.64619235914398,
+    "buffer_seconds": 60.22392341930631,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 363627.72883542767,
-    "upload_bytes_per_second": 4764598.978973576,
+    "generation_bytes_per_second": 314095.54621762695,
+    "upload_bytes_per_second": 4777199.77405335,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.02604578778384,
+    "estimated_full_shards_per_hour": 67.39759244820219,
     "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": 1.3138526910006476,
-      "bytes": 6032843,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.2020050520004588,
       "bytes": 6391888,
@@ -562,6 +556,12 @@
       "bytes": 932363,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.3606180970018613,
+      "bytes": 7382890,
+      "pages": 28,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -251,3 +251,4 @@
 {"step": 950100, "loss": 0.5853739976882935, "lr": 3e-05, "grad_norm": 0.35853561981526255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139068560747, "tokens_per_second": 145432.7120355331, "data_wait_s": 1.4164997992338613e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 950200, "loss": 0.5775582790374756, "lr": 3e-05, "grad_norm": 0.3584976870135802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139083407068, "tokens_per_second": 154612.7728279286, "data_wait_s": 1.4946999726817012e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 950300, "loss": 0.5320939421653748, "lr": 3e-05, "grad_norm": 0.34095561353472936, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139098139897, "tokens_per_second": 143550.26608170656, "data_wait_s": 1.6551999578950927e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 950400, "loss": 0.4987389147281647, "lr": 3e-05, "grad_norm": 0.33573880354791236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139112727748, "tokens_per_second": 151723.88885000825, "data_wait_s": 1.0514999303268269e-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:e3c5a7d53d532112339215063e8062a01af0347bae0b14d1a536bb11af3f11d3
-size 539934
+oid sha256:f2464b424e13b1e0838c097c8906f94323aa209733fe0e8cacfbbcc176ee7932
+size 542052