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": 74645.44109816219,
-      "required_pages_per_second_upper_estimate": 0.0064700776472478684
+      "required_token_positions_per_second": 76268.15367867655,
+      "required_pages_per_second_upper_estimate": 0.006610730260999418
     },
     {
       "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": 9330.680137270274,
-      "required_pages_per_second_upper_estimate": 0.0037221478128571382
+      "required_token_positions_per_second": 9533.519209834569,
+      "required_pages_per_second_upper_estimate": 0.003803063351617428
     },
     {
       "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": 18661.360274540548,
-      "required_pages_per_second_upper_estimate": 0.004316977270791358
+      "required_token_positions_per_second": 19067.038419669138,
+      "required_pages_per_second_upper_estimate": 0.004410823769975353
     },
     {
       "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": 27992.040411810824,
-      "required_pages_per_second_upper_estimate": 0.0017045683981934002
+      "required_token_positions_per_second": 28600.557629503706,
+      "required_pages_per_second_upper_estimate": 0.001741623904107796
     },
     {
       "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": 27992.040411810824,
-      "required_pages_per_second_upper_estimate": 0.002154904757828441
+      "required_token_positions_per_second": 28600.557629503706,
+      "required_pages_per_second_upper_estimate": 0.0022017500977299094
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27992.040411810824,
-      "required_pages_per_second_upper_estimate": 0.34756745858190424
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28600.557629503706,
+      "required_pages_per_second_upper_estimate": 0.35512320584953133
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08677191897638181,
-    "producer_p95_seconds": 0.25783332500213874,
-    "gpu_update_seconds": 0.987709348559469,
-    "producer_batches_per_second": 11.524465654288306,
-    "gpu_batches_per_second": 1.012443591283667,
-    "producer_headroom": 11.38282246389308,
+    "producer_batch_seconds": 0.0891979150782447,
+    "producer_p95_seconds": 0.2729030230002536,
+    "gpu_update_seconds": 0.9666944385545452,
+    "producer_batches_per_second": 11.21102437341497,
+    "gpu_batches_per_second": 1.0344530392615634,
+    "producer_headroom": 10.837634912279706,
     "sustainable": true,
-    "buffer_seconds": 60.25027026212761,
+    "buffer_seconds": 60.901749628936344,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311248.31757857127,
-    "upload_bytes_per_second": 5188847.819482128,
+    "generation_bytes_per_second": 311093.45059549477,
+    "upload_bytes_per_second": 5302327.769133626,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.78664346235135,
+    "estimated_full_shards_per_hour": 66.753412612902,
     "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": 2.1342484679989866,
-      "bytes": 1010482,
-      "pages": 32,
-      "shards": 1
-    },
     {
       "seconds": 1.0687636729999213,
       "bytes": 11353142,
@@ -562,6 +556,12 @@
       "bytes": 6964732,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.278426272001525,
+      "bytes": 5758474,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -184,3 +184,4 @@
 {"step": 173400, "loss": 0.6181184854358435, "lr": 0.0003, "grad_norm": 0.14915258734241732, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25378081454, "tokens_per_second": 148120.81649753303, "data_wait_s": 9.430998034076765e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 173500, "loss": 0.5983242467045784, "lr": 0.0003, "grad_norm": 0.1514536309798863, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25392879686, "tokens_per_second": 150984.8431385962, "data_wait_s": 1.2137999874539673e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 173600, "loss": 0.630302183330059, "lr": 0.0003, "grad_norm": 0.14655039917897708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25407624867, "tokens_per_second": 151793.06158211792, "data_wait_s": 9.817998943617567e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 173700, "loss": 0.5771032311022282, "lr": 0.0003, "grad_norm": 0.15708337584257606, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25422283775, "tokens_per_second": 136087.01565575678, "data_wait_s": 1.3551998563343659e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5dbd2c247d69a697d0e2284cf485b7253b3f0342f8db768814d10e0f0ba498c1
-size 396462
+oid sha256:9b17c5b192f4001c4ef198abbb08f5211d688da3df44347496d23310ad6caa86
+size 398580