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": 77087.69325172872,
-      "required_pages_per_second_upper_estimate": 0.005333105487246151
+      "required_token_positions_per_second": 76444.99336729341,
+      "required_pages_per_second_upper_estimate": 0.005288642017971727
     },
     {
       "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": 9635.96165646609,
-      "required_pages_per_second_upper_estimate": 0.004034375923801276
+      "required_token_positions_per_second": 9555.624170911677,
+      "required_pages_per_second_upper_estimate": 0.004000740296237111
     },
     {
       "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": 19271.92331293218,
-      "required_pages_per_second_upper_estimate": 0.004013833391288702
+      "required_token_positions_per_second": 19111.248341823353,
+      "required_pages_per_second_upper_estimate": 0.003980369032090666
     },
     {
       "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": 28907.88496939827,
-      "required_pages_per_second_upper_estimate": 0.0017329835641308544
+      "required_token_positions_per_second": 28666.87251273503,
+      "required_pages_per_second_upper_estimate": 0.0017185352353586063
     },
     {
       "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": 28907.88496939827,
-      "required_pages_per_second_upper_estimate": 0.0022474609040226095
+      "required_token_positions_per_second": 28666.87251273503,
+      "required_pages_per_second_upper_estimate": 0.0022287232456188045
     },
     {
       "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": 28907.88496939827,
-      "required_pages_per_second_upper_estimate": 0.3707993095188398
+      "required_token_positions_per_second": 28666.87251273503,
+      "required_pages_per_second_upper_estimate": 0.3677078605037779
     }
   ],
   "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": 0.12350419612494079,
-    "producer_p95_seconds": 0.25509821599916904,
-    "gpu_update_seconds": 0.9564172553359758,
-    "producer_batches_per_second": 8.096890886106964,
-    "gpu_batches_per_second": 1.0455687561269629,
-    "producer_headroom": 7.744006158045298,
+    "producer_batch_seconds": 0.07458110692186892,
+    "producer_p95_seconds": 0.2620379159998265,
+    "gpu_update_seconds": 0.9644581908165112,
+    "producer_batches_per_second": 13.408221482252856,
+    "gpu_batches_per_second": 1.0368515810451038,
+    "producer_headroom": 12.931669032840668,
     "sustainable": true,
-    "buffer_seconds": 60.25428708616647,
+    "buffer_seconds": 60.76086602144021,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 389952.99835796416,
-    "upload_bytes_per_second": 4803024.947133813,
+    "generation_bytes_per_second": 389704.2455578036,
+    "upload_bytes_per_second": 4775112.971789076,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.67483580640977,
+    "estimated_full_shards_per_hour": 83.62145924616414,
     "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": 1.515185850999842,
-      "bytes": 11700777,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.3114747930012527,
       "bytes": 6594730,
@@ -562,6 +556,12 @@
       "bytes": 887856,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.2455697520017566,
+      "bytes": 7910132,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -241,3 +241,4 @@
 {"step": 949100, "loss": 0.5535507202148438, "lr": 3e-05, "grad_norm": 0.37402986916779357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138921774854, "tokens_per_second": 153016.35622632026, "data_wait_s": 1.4047000149730593e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 949200, "loss": 0.5632185339927673, "lr": 3e-05, "grad_norm": 0.35542638318195346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138936514089, "tokens_per_second": 153708.60155259547, "data_wait_s": 1.0625997674651444e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 949300, "loss": 0.5321933031082153, "lr": 3e-05, "grad_norm": 0.3530456270577749, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138951280039, "tokens_per_second": 153581.8055451676, "data_wait_s": 8.073999197222292e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 949400, "loss": 0.4543458819389343, "lr": 3e-05, "grad_norm": 0.3367170802597107, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138966051757, "tokens_per_second": 154100.82524088165, "data_wait_s": 8.947998139774427e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bccd40d3a736ecdd66c2e709b24d8a0282387901f2326d28a53cc2b785f99aad
-size 517971
+oid sha256:505386530d0e7e5d8a189114517486723da79f53846349077ece0e182b584649
+size 520089