Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75286.70000909828,
-      "required_pages_per_second_upper_estimate": 0.006525660344391958
+      "required_token_positions_per_second": 75703.71619841477,
+      "required_pages_per_second_upper_estimate": 0.006561806250764041
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9410.837501137285,
-      "required_pages_per_second_upper_estimate": 0.003754123783763078
+      "required_token_positions_per_second": 9462.964524801846,
+      "required_pages_per_second_upper_estimate": 0.0037749180328713284
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18821.67500227457,
-      "required_pages_per_second_upper_estimate": 0.00435406326160978
+      "required_token_positions_per_second": 18925.929049603692,
+      "required_pages_per_second_upper_estimate": 0.004378180600650806
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28232.512503411854,
-      "required_pages_per_second_upper_estimate": 0.0017192118869123446
+      "required_token_positions_per_second": 28388.893574405538,
+      "required_pages_per_second_upper_estimate": 0.0017287346736677895
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28232.512503411854,
-      "required_pages_per_second_upper_estimate": 0.0021734169651092426
+      "required_token_positions_per_second": 28388.893574405538,
+      "required_pages_per_second_upper_estimate": 0.002185455612844847
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28232.512503411854,
-      "required_pages_per_second_upper_estimate": 0.35055331715127025
+      "required_token_positions_per_second": 28388.893574405538,
+      "required_pages_per_second_upper_estimate": 0.35249504667923487
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.15289794686771074,
-    "producer_p95_seconds": 0.4431833850030671,
-    "gpu_update_seconds": 0.9792964758860475,
-    "producer_batches_per_second": 6.540310190464576,
-    "gpu_batches_per_second": 1.0211412219115976,
-    "producer_headroom": 6.404902720723564,
+    "producer_batch_seconds": 0.2894173106877247,
+    "producer_p95_seconds": 0.4313449370019953,
+    "gpu_update_seconds": 0.9739019919017379,
+    "producer_batches_per_second": 3.4552183406851547,
+    "gpu_batches_per_second": 1.0267973659724225,
+    "producer_headroom": 3.36504402444869,
     "sustainable": true,
-    "buffer_seconds": 60.716381504934944,
+    "buffer_seconds": 60.38192349790775,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 170,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 579026.0880397903,
-    "upload_bytes_per_second": 5770932.083898797,
+    "generation_bytes_per_second": 538605.0892061491,
+    "upload_bytes_per_second": 5778877.514367516,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.12275972793236,
+    "estimated_full_shards_per_hour": 115.57211405886035,
     "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": 1.1455287000062526,
-      "bytes": 6484150,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.8978742559993407,
       "bytes": 2461928,
@@ -562,6 +556,12 @@
       "bytes": 1242871,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.21954084400204,
+      "bytes": 7471926,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -416,3 +416,4 @@
 {"step": 156600, "loss": 0.6383127085864544, "lr": 0.0003, "grad_norm": 0.1486443915818143, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22918824594, "tokens_per_second": 148071.14429043987, "data_wait_s": 7.271002687048167e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 156700, "loss": 0.6271420363336802, "lr": 0.0003, "grad_norm": 0.14207540156982212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22933594966, "tokens_per_second": 149792.27462428136, "data_wait_s": 1.5994002751540393e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 156800, "loss": 0.61750128865242, "lr": 0.0003, "grad_norm": 0.15084095647069573, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22948289405, "tokens_per_second": 149668.41289874725, "data_wait_s": 1.2961994798388332e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 156900, "loss": 0.5993822142481804, "lr": 0.0003, "grad_norm": 0.14820837760259123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22963019955, "tokens_per_second": 150153.0592304569, "data_wait_s": 9.803996363189071e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:de11a0dbd8ca631f8fff94c4580e60f202ce2c2b05d406caee5ae977fab0d5ed
-size 891753
+oid sha256:c04a34888ea236f3d31688790d9cd40cb3e19a356f913077a94cf51de0b13286
+size 893871