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": 75843.50955116315,
-      "required_pages_per_second_upper_estimate": 0.006573923184277258
+      "required_token_positions_per_second": 73687.43018590803,
+      "required_pages_per_second_upper_estimate": 0.006387039689430136
     },
     {
       "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": 9480.438693895394,
-      "required_pages_per_second_upper_estimate": 0.003781888740184855
+      "required_token_positions_per_second": 9210.928773238504,
+      "required_pages_per_second_upper_estimate": 0.003674377203302419
     },
     {
       "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": 18960.877387790788,
-      "required_pages_per_second_upper_estimate": 0.0043862652836206455
+      "required_token_positions_per_second": 18421.857546477007,
+      "required_pages_per_second_upper_estimate": 0.004261572529757908
     },
     {
       "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": 28441.316081686182,
-      "required_pages_per_second_upper_estimate": 0.0017319269293215411
+      "required_token_positions_per_second": 27632.78631971551,
+      "required_pages_per_second_upper_estimate": 0.001682691708845347
     },
     {
       "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": 28441.316081686182,
-      "required_pages_per_second_upper_estimate": 0.002189491242570099
+      "required_token_positions_per_second": 27632.78631971551,
+      "required_pages_per_second_upper_estimate": 0.0021272483833399674
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28441.316081686182,
-      "required_pages_per_second_upper_estimate": 0.35314595877281474
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27632.78631971551,
+      "required_pages_per_second_upper_estimate": 0.34310672510418205
     }
   ],
   "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": 62,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.1543300060702677,
-    "producer_p95_seconds": 0.2544935459991393,
-    "gpu_update_seconds": 0.9721069137796682,
-    "producer_batches_per_second": 6.479621335235948,
-    "gpu_batches_per_second": 1.028693434667469,
-    "producer_headroom": 6.29888469865711,
+    "producer_batch_seconds": 0.12470503284367851,
+    "producer_p95_seconds": 0.3162095070001669,
+    "gpu_update_seconds": 1.0005505662769025,
+    "producer_batches_per_second": 8.018922550251279,
+    "gpu_batches_per_second": 0.9994497366795252,
+    "producer_headroom": 8.02333749858454,
     "sustainable": true,
-    "buffer_seconds": 60.270628654339426,
+    "buffer_seconds": 60.03303397661415,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 393912.24994001404,
-    "upload_bytes_per_second": 5490991.8692655815,
+    "generation_bytes_per_second": 353835.4745118726,
+    "upload_bytes_per_second": 5444500.8166638585,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.52439902925792,
+    "estimated_full_shards_per_hour": 75.92485596196302,
     "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": 1.2206816289999551,
-      "bytes": 9827767,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.1708706409990555,
       "bytes": 7164779,
@@ -562,6 +556,12 @@
       "bytes": 6275188,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9176649480032211,
+      "bytes": 4600807,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -209,3 +209,4 @@
 {"step": 258000, "loss": 0.585047347471118, "lr": 0.0003, "grad_norm": 0.15418200421276249, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37764114907, "tokens_per_second": 150328.98359405634, "data_wait_s": 1.2985998182557523e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 258100, "loss": 0.6123698111623526, "lr": 0.0003, "grad_norm": 0.15422787591436724, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37778839572, "tokens_per_second": 148994.77597158717, "data_wait_s": 1.5338999219238758e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 258200, "loss": 0.5067947395145893, "lr": 0.0003, "grad_norm": 0.15570042932146388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37793466222, "tokens_per_second": 148696.57369766242, "data_wait_s": 1.4452001778408885e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 258300, "loss": 0.596388328820467, "lr": 0.0003, "grad_norm": 0.14077683984788145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37807949727, "tokens_per_second": 146497.8689961445, "data_wait_s": 9.57899828790687e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7790e9d9ce02910da144e1ed2972923fd74dba988fe573910167c33fe361b0f1
-size 450195
+oid sha256:566eab8da38dccebc4a10265ae4dd442b8ebe6b7e9e7fee92512c39eb94c8f12
+size 452313