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": 77182.09558042981,
-      "required_pages_per_second_upper_estimate": 0.005339636459389273
+      "required_token_positions_per_second": 77249.75229550217,
+      "required_pages_per_second_upper_estimate": 0.005344317107923186
     },
     {
       "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": 9647.761947553727,
-      "required_pages_per_second_upper_estimate": 0.00403931645926157
+      "required_token_positions_per_second": 9656.219036937771,
+      "required_pages_per_second_upper_estimate": 0.004042857265982561
     },
     {
       "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": 19295.523895107453,
-      "required_pages_per_second_upper_estimate": 0.004018748770166605
+      "required_token_positions_per_second": 19312.438073875543,
+      "required_pages_per_second_upper_estimate": 0.0040222715475471075
     },
     {
       "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": 28943.285842661182,
-      "required_pages_per_second_upper_estimate": 0.0017351057924289623
+      "required_token_positions_per_second": 28968.657110813314,
+      "required_pages_per_second_upper_estimate": 0.0017366267612149992
     },
     {
       "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": 28943.285842661182,
-      "required_pages_per_second_upper_estimate": 0.0022502131662068157
+      "required_token_positions_per_second": 28968.657110813314,
+      "required_pages_per_second_upper_estimate": 0.002252185670709229
     },
     {
       "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": 28943.285842661182,
-      "required_pages_per_second_upper_estimate": 0.37125339391056017
+      "required_token_positions_per_second": 28968.657110813314,
+      "required_pages_per_second_upper_estimate": 0.3715788292968704
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.09229515982832481,
-    "producer_p95_seconds": 0.26921247399877757,
-    "gpu_update_seconds": 0.95524745014431,
-    "producer_batches_per_second": 10.834804358755834,
-    "gpu_batches_per_second": 1.0468491696564375,
-    "producer_headroom": 10.349919236513966,
+    "producer_batch_seconds": 0.46538089410171324,
+    "producer_p95_seconds": 0.2749621870025294,
+    "gpu_update_seconds": 0.954410827337925,
+    "producer_batches_per_second": 2.148777512515245,
+    "gpu_batches_per_second": 1.0477668225843937,
+    "producer_headroom": 2.0508165234848037,
     "sustainable": true,
-    "buffer_seconds": 60.18058935909153,
+    "buffer_seconds": 60.12788212228928,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 351347.98020013707,
-    "upload_bytes_per_second": 4949350.682889696,
+    "generation_bytes_per_second": 320862.72648135596,
+    "upload_bytes_per_second": 4894699.368225426,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.39109758856854,
+    "estimated_full_shards_per_hour": 68.84967180090436,
     "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": 1.4462591800001974,
-      "bytes": 10549433,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.0944702749984572,
       "bytes": 2560030,
@@ -562,6 +556,12 @@
       "bytes": 6068190,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.6203178860014305,
+      "bytes": 6721657,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -243,3 +243,4 @@
 {"step": 866800, "loss": 0.5601211786270142, "lr": 3e-05, "grad_norm": 0.30758305744515335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126875642561, "tokens_per_second": 152442.34845604206, "data_wait_s": 1.5159002941800281e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 866900, "loss": 0.5435276627540588, "lr": 3e-05, "grad_norm": 0.3072318468543782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126890142047, "tokens_per_second": 150199.6450129293, "data_wait_s": 1.091699959943071e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 867000, "loss": 0.5461263060569763, "lr": 3e-05, "grad_norm": 0.2977331601759649, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126904696431, "tokens_per_second": 128672.02598314956, "data_wait_s": 1.4255998394219205e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 867100, "loss": 0.5442878603935242, "lr": 3e-05, "grad_norm": 0.31072423824450185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126919416590, "tokens_per_second": 152392.37245601066, "data_wait_s": 1.6568999853916466e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:945559c4bd8387d911fb2332caf6c102179ad1e46d231904651ce2815c3dcccd
-size 522990
+oid sha256:2842d1a18d9b35ede368927a6b039d5ec8fb61976b790964a1012ac6aee80b53
+size 525108