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": 76919.94777012474,
-      "required_pages_per_second_upper_estimate": 0.005321500465605645
+      "required_token_positions_per_second": 76285.71196533572,
+      "required_pages_per_second_upper_estimate": 0.005277622561000523
     },
     {
       "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": 9614.993471265592,
-      "required_pages_per_second_upper_estimate": 0.004025596982523326
+      "required_token_positions_per_second": 9535.713995666965,
+      "required_pages_per_second_upper_estimate": 0.003992404321633889
     },
     {
       "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": 19229.986942531184,
-      "required_pages_per_second_upper_estimate": 0.004005099151270632
+      "required_token_positions_per_second": 19071.42799133393,
+      "required_pages_per_second_upper_estimate": 0.00397207550321698
     },
     {
       "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": 28844.980413796777,
-      "required_pages_per_second_upper_estimate": 0.00172921253207224
+      "required_token_positions_per_second": 28607.141987000894,
+      "required_pages_per_second_upper_estimate": 0.0017149544815440768
     },
     {
       "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": 28844.980413796777,
-      "required_pages_per_second_upper_estimate": 0.002242570351512492
+      "required_token_positions_per_second": 28607.141987000894,
+      "required_pages_per_second_upper_estimate": 0.002224079459969792
     },
     {
       "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": 28844.980413796777,
-      "required_pages_per_second_upper_estimate": 0.3699924374212334
+      "required_token_positions_per_second": 28607.141987000894,
+      "required_pages_per_second_upper_estimate": 0.36694170145330224
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08993021281250435,
-    "producer_p95_seconds": 0.26975980500355945,
-    "gpu_update_seconds": 0.9585029909320288,
-    "producer_batches_per_second": 11.119733499184548,
-    "gpu_batches_per_second": 1.0432935624203117,
-    "producer_headroom": 10.658297817335463,
+    "producer_batch_seconds": 0.21675075976563107,
+    "producer_p95_seconds": 0.2752362049977819,
+    "gpu_update_seconds": 0.9664719395094858,
+    "producer_batches_per_second": 4.613593978084705,
+    "gpu_batches_per_second": 1.0346911887659467,
+    "producer_headroom": 4.4589091201088085,
     "sustainable": true,
-    "buffer_seconds": 60.38568842871782,
+    "buffer_seconds": 60.887732189097605,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 169,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 505934.9399003648,
-    "upload_bytes_per_second": 4787220.251045446,
+    "generation_bytes_per_second": 614227.4902624867,
+    "upload_bytes_per_second": 4817176.597645194,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.56186053999146,
+    "estimated_full_shards_per_hour": 65.89946046307541,
     "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": 1.1808424570008356,
-      "bytes": 2620310,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1430494310006907,
       "bytes": 10900083,
@@ -562,6 +556,12 @@
       "bytes": 2585533,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3021024750014476,
+      "bytes": 5911875,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -245,3 +245,4 @@
 {"step": 464700, "loss": 0.5485379695892334, "lr": 0.0003, "grad_norm": 0.1977374657831521, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68017987997, "tokens_per_second": 152362.4185223103, "data_wait_s": 1.6730002244003117e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 464800, "loss": 0.5927941799163818, "lr": 0.0003, "grad_norm": 0.1538716172456637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68032688293, "tokens_per_second": 152428.54795846582, "data_wait_s": 8.838000212563202e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 464900, "loss": 0.5823112726211548, "lr": 0.0003, "grad_norm": 0.1557854975686474, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68047434277, "tokens_per_second": 153518.83346817302, "data_wait_s": 1.333300315309316e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 465000, "loss": 0.6226056218147278, "lr": 0.0003, "grad_norm": 0.1717177493832919, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68062086371, "tokens_per_second": 151632.5618039361, "data_wait_s": 1.2479998986236751e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ef2dced8eeef27494145de6aea9815fcbf523045b45ea8cbe9e35f5bc90bfdb2
-size 526443
+oid sha256:87c48e401f9b5c9b963eeb3f097aace0f60f41e12aeffecf6bca2995407452ee
+size 528561