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": 77165.46200678668,
-      "required_pages_per_second_upper_estimate": 0.005338485710169433
+      "required_token_positions_per_second": 76526.14742804598,
+      "required_pages_per_second_upper_estimate": 0.005294256444197954
     },
     {
       "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": 9645.682750848335,
-      "required_pages_per_second_upper_estimate": 0.004038445943019587
+      "required_token_positions_per_second": 9565.768428505748,
+      "required_pages_per_second_upper_estimate": 0.0040049874850555615
     },
     {
       "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": 19291.36550169667,
-      "required_pages_per_second_upper_estimate": 0.004017882686483348
+      "required_token_positions_per_second": 19131.536857011495,
+      "required_pages_per_second_upper_estimate": 0.00398459459476023
     },
     {
       "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": 28937.048252545006,
-      "required_pages_per_second_upper_estimate": 0.0017347318584257463
+      "required_token_positions_per_second": 28697.305285517243,
+      "required_pages_per_second_upper_estimate": 0.0017203596336187465
     },
     {
       "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": 28937.048252545006,
-      "required_pages_per_second_upper_estimate": 0.0022497282210115423
+      "required_token_positions_per_second": 28697.305285517243,
+      "required_pages_per_second_upper_estimate": 0.0022310892598430008
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28937.048252545006,
-      "required_pages_per_second_upper_estimate": 0.37117338480195233
+      "required_token_positions_per_second": 28697.305285517243,
+      "required_pages_per_second_upper_estimate": 0.36809821943686255
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.2789821175387317,
-    "producer_p95_seconds": 0.23711803700280143,
-    "gpu_update_seconds": 0.9554533606435952,
-    "producer_batches_per_second": 3.5844591360274833,
-    "gpu_batches_per_second": 1.0466235623750364,
-    "producer_headroom": 3.424783527607097,
+    "producer_batch_seconds": 0.21171202975781966,
+    "producer_p95_seconds": 0.2590065639997192,
+    "gpu_update_seconds": 0.9634354070851803,
+    "producer_batches_per_second": 4.723397159546928,
+    "gpu_batches_per_second": 1.0379523034402938,
+    "producer_headroom": 4.550688065233079,
     "sustainable": true,
-    "buffer_seconds": 60.1935617205465,
+    "buffer_seconds": 60.69643064636636,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 493199.22826282814,
-    "upload_bytes_per_second": 3773051.077990002,
+    "generation_bytes_per_second": 485006.1658383242,
+    "upload_bytes_per_second": 3724219.4579773345,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.8290732947696,
+    "estimated_full_shards_per_hour": 104.07103282320303,
     "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": 1.290630097999383,
-      "bytes": 12177821,
-      "pages": 42,
-      "shards": 3
-    },
     {
       "seconds": 3.9100393570006418,
       "bytes": 2234803,
@@ -562,6 +556,12 @@
       "bytes": 7866146,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.004612441000063,
+      "bytes": 5431394,
+      "pages": 26,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -242,3 +242,4 @@
 {"step": 549200, "loss": 0.5703690052032471, "lr": 0.0003, "grad_norm": 0.14491462774986033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80385500072, "tokens_per_second": 151527.3272872314, "data_wait_s": 1.442299981135875e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 549300, "loss": 0.5789375305175781, "lr": 0.0003, "grad_norm": 0.1926010809249785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80399965945, "tokens_per_second": 151181.70045628812, "data_wait_s": 1.2969998351763934e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 549400, "loss": 0.6239665150642395, "lr": 0.0003, "grad_norm": 0.18234078992583755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80414638023, "tokens_per_second": 152883.3989896266, "data_wait_s": 1.0659998224582523e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 549500, "loss": 0.6144766211509705, "lr": 0.0003, "grad_norm": 0.17061365900134132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80429278828, "tokens_per_second": 153154.0853361587, "data_wait_s": 7.97999746282585e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:24948941a1eae4797a518df13e534d31b17f856c5ff594760ad532ecdd26f55f
-size 520089
+oid sha256:42f9ace6170ca55b19766bd113ff739072d488c037c2d1813c8daf8fbba7960a
+size 522207