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": 77415.77643470117,
-      "required_pages_per_second_upper_estimate": 0.00535580304310205
+      "required_token_positions_per_second": 75359.31795997874,
+      "required_pages_per_second_upper_estimate": 0.005213532474179688
     },
     {
       "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": 9676.972054337646,
-      "required_pages_per_second_upper_estimate": 0.004051546120995605
+      "required_token_positions_per_second": 9419.914744997342,
+      "required_pages_per_second_upper_estimate": 0.003943921593541851
     },
     {
       "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": 19353.944108675292,
-      "required_pages_per_second_upper_estimate": 0.00403091616000815
+      "required_token_positions_per_second": 18839.829489994685,
+      "required_pages_per_second_upper_estimate": 0.003923839643051218
     },
     {
       "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": 29030.916163012942,
-      "required_pages_per_second_upper_estimate": 0.0017403590963302987
+      "required_token_positions_per_second": 28259.744234992027,
+      "required_pages_per_second_upper_estimate": 0.0016941285167567989
     },
     {
       "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": 29030.916163012942,
-      "required_pages_per_second_upper_estimate": 0.0022570260381691207
+      "required_token_positions_per_second": 28259.744234992027,
+      "required_pages_per_second_upper_estimate": 0.0021970708127923254
     },
     {
       "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": 29030.916163012942,
-      "required_pages_per_second_upper_estimate": 0.37237742156992526
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28259.744234992027,
+      "required_pages_per_second_upper_estimate": 0.36248565609717714
     }
   ],
   "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": 64,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5890331506637381,
-    "producer_p95_seconds": 0.3250934009993216,
-    "gpu_update_seconds": 0.9523640192666446,
-    "producer_batches_per_second": 1.697697317838858,
-    "gpu_batches_per_second": 1.0500186690904565,
-    "producer_headroom": 1.616825841115217,
+    "producer_batch_seconds": 0.15691575543735325,
+    "producer_p95_seconds": 0.33196828899963293,
+    "gpu_update_seconds": 0.9783528035531706,
+    "producer_batches_per_second": 6.372846354483747,
+    "gpu_batches_per_second": 1.0221261659068297,
+    "producer_headroom": 6.234892097522777,
     "sustainable": true,
-    "buffer_seconds": 60.951297233065254,
+    "buffer_seconds": 60.65787382029657,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 243914.183979659,
-    "upload_bytes_per_second": 3830068.502272294,
+    "generation_bytes_per_second": 244543.61852903746,
+    "upload_bytes_per_second": 3864936.90270717,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.338305850432654,
+    "estimated_full_shards_per_hour": 52.47336785224288,
     "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": 1.0856028650014196,
-      "bytes": 1171092,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 3.5694688390030933,
       "bytes": 7359997,
@@ -562,6 +556,12 @@
       "bytes": 11305656,
       "pages": 28,
       "shards": 3
+    },
+    {
+      "seconds": 1.8134901779994834,
+      "bytes": 7862983,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -364,3 +364,4 @@
 {"step": 878900, "loss": 0.539548397064209, "lr": 3e-05, "grad_norm": 0.32667433365152343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128646391144, "tokens_per_second": 151418.38823903332, "data_wait_s": 1.35770023916848e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 879000, "loss": 0.560137152671814, "lr": 3e-05, "grad_norm": 0.3201839729736884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128661075387, "tokens_per_second": 128963.10808562308, "data_wait_s": 1.383299968438223e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 879100, "loss": 0.526741623878479, "lr": 3e-05, "grad_norm": 0.3095050575443668, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128675793684, "tokens_per_second": 151964.4515750159, "data_wait_s": 1.0734998795669526e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 879200, "loss": 0.5450117588043213, "lr": 3e-05, "grad_norm": 0.3068843426753608, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128690530526, "tokens_per_second": 153357.47921515413, "data_wait_s": 8.413000614382327e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d041371c9759381555f8a432e80adc7be2acd4db4e8b7ac53db36d949665c7ee
-size 780834
+oid sha256:492c9520386bb485936e23b7f3dda453a10a35f95ffdc0f4df7e5e927e0877a7
+size 782952