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": 75505.81804772685,
-      "required_pages_per_second_upper_estimate": 0.005223667690177138
+      "required_token_positions_per_second": 76279.88003806896,
+      "required_pages_per_second_upper_estimate": 0.0052772190947402734
     },
     {
       "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": 9438.227255965856,
-      "required_pages_per_second_upper_estimate": 0.003951588659475653
+      "required_token_positions_per_second": 9534.98500475862,
+      "required_pages_per_second_upper_estimate": 0.003992099108363626
     },
     {
       "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": 18876.45451193171,
-      "required_pages_per_second_upper_estimate": 0.003931467669253921
+      "required_token_positions_per_second": 19069.97000951724,
+      "required_pages_per_second_upper_estimate": 0.003971771844054129
     },
     {
       "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": 28314.681767897568,
-      "required_pages_per_second_upper_estimate": 0.0016974219379697303
+      "required_token_positions_per_second": 28604.95501427586,
+      "required_pages_per_second_upper_estimate": 0.001714823375868525
     },
     {
       "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": 28314.681767897568,
-      "required_pages_per_second_upper_estimate": 0.002201341964331062
+      "required_token_positions_per_second": 28604.95501427586,
+      "required_pages_per_second_upper_estimate": 0.0022239094324598974
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28314.681767897568,
-      "required_pages_per_second_upper_estimate": 0.36319033578196236
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28604.95501427586,
+      "required_pages_per_second_upper_estimate": 0.36691364931537385
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.13046524485926625,
-    "producer_p95_seconds": 0.2754479169998376,
-    "gpu_update_seconds": 0.9764545555071915,
-    "producer_batches_per_second": 7.664876581335564,
-    "gpu_batches_per_second": 1.0241132005171285,
-    "producer_headroom": 7.4844036552455,
+    "producer_batch_seconds": 0.0793257537345653,
+    "producer_p95_seconds": 0.27102510900294874,
+    "gpu_update_seconds": 0.9665458304759342,
+    "producer_batches_per_second": 12.606246432225975,
+    "gpu_batches_per_second": 1.0346120881899543,
+    "producer_headroom": 12.184514927020137,
     "sustainable": true,
-    "buffer_seconds": 60.54018244144587,
+    "buffer_seconds": 60.892387319983854,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 171,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 532169.0416109266,
-    "upload_bytes_per_second": 5265012.031448488,
+    "generation_bytes_per_second": 702722.5693498176,
+    "upload_bytes_per_second": 5134699.5794174895,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.19108806844568,
+    "estimated_full_shards_per_hour": 75.39395241914222,
     "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": 1.3733238169988908,
-      "bytes": 13560061,
-      "pages": 41,
-      "shards": 4
-    },
     {
       "seconds": 1.2763601959995867,
       "bytes": 5814992,
@@ -562,6 +556,12 @@
       "bytes": 2772365,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.9406323449984484,
+      "bytes": 5788082,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -316,3 +316,4 @@
 {"step": 751600, "loss": 0.4991948902606964, "lr": 5e-05, "grad_norm": 0.25065669006562635, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110013115576, "tokens_per_second": 151425.5367453049, "data_wait_s": 1.6547997802263126e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 751700, "loss": 0.5572493076324463, "lr": 5e-05, "grad_norm": 0.24770711443238044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110027801931, "tokens_per_second": 151018.90590762708, "data_wait_s": 1.3755001418758184e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 751800, "loss": 0.5707524418830872, "lr": 5e-05, "grad_norm": 0.25372712582124146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110042453269, "tokens_per_second": 151336.983288377, "data_wait_s": 1.3872999261366203e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 751900, "loss": 0.4962722361087799, "lr": 5e-05, "grad_norm": 0.24667855449304538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110056944555, "tokens_per_second": 149582.60326599845, "data_wait_s": 1.2260999938007444e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:01feb252266c1ecc2f4676ba665257e9fb2eb2493e8df10024bb0d4248e3db5a
-size 678387
+oid sha256:c0b84fef6b4b1178c363dd7244ed23eed7e1553d6942eb55e41474d9500e6376
+size 680505