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": 75674.31483945703,
-      "required_pages_per_second_upper_estimate": 0.0065592578154866485
+      "required_token_positions_per_second": 74696.70782709865,
+      "required_pages_per_second_upper_estimate": 0.006474521317377744
     },
     {
       "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": 9459.289354932129,
-      "required_pages_per_second_upper_estimate": 0.003773451952661612
+      "required_token_positions_per_second": 9337.088478387332,
+      "required_pages_per_second_upper_estimate": 0.0037247041959419707
     },
     {
       "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": 18918.578709864258,
-      "required_pages_per_second_upper_estimate": 0.004376480228913649
+      "required_token_positions_per_second": 18674.176956774663,
+      "required_pages_per_second_upper_estimate": 0.004319942184660317
     },
     {
       "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": 28377.86806479639,
-      "required_pages_per_second_upper_estimate": 0.0017280632779789705
+      "required_token_positions_per_second": 28011.265435161993,
+      "required_pages_per_second_upper_estimate": 0.0017057391012495842
     },
     {
       "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": 28377.86806479639,
-      "required_pages_per_second_upper_estimate": 0.002184606838594571
+      "required_token_positions_per_second": 28011.265435161993,
+      "required_pages_per_second_upper_estimate": 0.002156384753344285
     },
     {
       "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": 28377.86806479639,
-      "required_pages_per_second_upper_estimate": 0.3523581467498962
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28011.265435161993,
+      "required_pages_per_second_upper_estimate": 0.3478061690299116
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07508743142184926,
-    "producer_p95_seconds": 0.20570114700058184,
-    "gpu_update_seconds": 0.9742803771188925,
-    "producer_batches_per_second": 13.317808068062583,
-    "gpu_batches_per_second": 1.0263985845195454,
-    "producer_headroom": 12.975279066949042,
+    "producer_batch_seconds": 0.3062706661561947,
+    "producer_p95_seconds": 0.22277083799963293,
+    "gpu_update_seconds": 0.9870314521847344,
+    "producer_batches_per_second": 3.265085790129215,
+    "gpu_batches_per_second": 1.0131389407972364,
+    "producer_headroom": 3.22274236893898,
     "sustainable": true,
-    "buffer_seconds": 60.40538338137134,
+    "buffer_seconds": 60.2089185832688,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 247838.88452880364,
-    "upload_bytes_per_second": 4955567.359769564,
+    "generation_bytes_per_second": 227954.08391677495,
+    "upload_bytes_per_second": 4969578.80704246,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.180455225926224,
+    "estimated_full_shards_per_hour": 48.91363990905224,
     "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": 0.9781999550000364,
-      "bytes": 5863999,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 0.9417934939999668,
       "bytes": 2584042,
@@ -562,6 +556,12 @@
       "bytes": 835967,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 1.0692436869994708,
+      "bytes": 7577178,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -78,3 +78,4 @@
 {"step": 7800, "loss": 0.7411314342170954, "lr": 0.0003, "grad_norm": 0.17838141708518138, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1140947358, "tokens_per_second": 148041.28679959124, "data_wait_s": 1.5087000065250322e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 7900, "loss": 0.7169801648706198, "lr": 0.0003, "grad_norm": 0.1563463019300498, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1155501400, "tokens_per_second": 140286.4032580151, "data_wait_s": 1.5573999917251058e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 8000, "loss": 0.7449607010930777, "lr": 0.0003, "grad_norm": 0.14968626283655082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1170105212, "tokens_per_second": 148722.72278340437, "data_wait_s": 1.4211000234354287e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 8100, "loss": 0.693482581526041, "lr": 0.0003, "grad_norm": 0.1611827912346656, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1184819071, "tokens_per_second": 149585.46369521818, "data_wait_s": 9.639001291361637e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b08d50122a114aa87f82e23a38e5f313cea65b7e602e0aa1daee0b36aa3da4a2
-size 167539
+oid sha256:c7cc4ccd4f112abdaf8b9b40a542fce1812757448c89b31d31c6a7dd935e9180
+size 169622