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": 75430.48184111103,
-      "required_pages_per_second_upper_estimate": 0.005218455756599373
+      "required_token_positions_per_second": 74181.2448506881,
+      "required_pages_per_second_upper_estimate": 0.005132030642972744
     },
     {
       "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": 9428.810230138879,
-      "required_pages_per_second_upper_estimate": 0.003947645947411759
+      "required_token_positions_per_second": 9272.655606336013,
+      "required_pages_per_second_upper_estimate": 0.0038822672673048506
     },
     {
       "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": 18857.620460277758,
-      "required_pages_per_second_upper_estimate": 0.003927545032981745
+      "required_token_positions_per_second": 18545.311212672026,
+      "required_pages_per_second_upper_estimate": 0.0038624992528541874
     },
     {
       "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": 28286.43069041664,
-      "required_pages_per_second_upper_estimate": 0.0016957283290116463
+      "required_token_positions_per_second": 27817.96681900804,
+      "required_pages_per_second_upper_estimate": 0.001667644635223615
     },
     {
       "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": 28286.43069041664,
-      "required_pages_per_second_upper_estimate": 0.002199145567320277
+      "required_token_positions_per_second": 27817.96681900804,
+      "required_pages_per_second_upper_estimate": 0.002162724561873063
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28286.43069041664,
-      "required_pages_per_second_upper_estimate": 0.36282796129368067
+      "required_token_positions_per_second": 27817.96681900804,
+      "required_pages_per_second_upper_estimate": 0.3568190097485671
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12512750606288137,
-    "producer_p95_seconds": 0.31889154400414554,
-    "gpu_update_seconds": 0.9774297896611983,
-    "producer_batches_per_second": 7.991847927484958,
-    "gpu_batches_per_second": 1.0230913878188888,
-    "producer_headroom": 7.8114702387659065,
+    "producer_batch_seconds": 0.11812832061724521,
+    "producer_p95_seconds": 0.33498035299999174,
+    "gpu_update_seconds": 0.9938900344473809,
+    "producer_batches_per_second": 8.465370495193621,
+    "gpu_batches_per_second": 1.0061475267291682,
+    "producer_headroom": 8.413647373077831,
     "sustainable": true,
-    "buffer_seconds": 60.6006469589943,
+    "buffer_seconds": 60.62729210129024,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 256034.5110494455,
-    "upload_bytes_per_second": 4712916.222313591,
+    "generation_bytes_per_second": 289167.635718674,
+    "upload_bytes_per_second": 4675890.7172043165,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.93904589283488,
+    "estimated_full_shards_per_hour": 62.04864314718404,
     "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": 1.7381740189994161,
-      "bytes": 7938212,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.2239695549978933,
       "bytes": 5214335,
@@ -562,6 +556,12 @@
       "bytes": 294229,
       "pages": 9,
       "shards": 1
+    },
+    {
+      "seconds": 1.2844244400039315,
+      "bytes": 2478394,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -400,3 +400,4 @@
 {"step": 760000, "loss": 0.5432776808738708, "lr": 5e-05, "grad_norm": 0.2574025637684899, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111241886638, "tokens_per_second": 150473.4611617272, "data_wait_s": 1.3263997971080244e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 760100, "loss": 0.5447261929512024, "lr": 5e-05, "grad_norm": 0.24335630870872557, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111256501409, "tokens_per_second": 135725.63104843546, "data_wait_s": 1.2349999451544136e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 760200, "loss": 0.5978831052780151, "lr": 5e-05, "grad_norm": 0.2611553275811896, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111271094700, "tokens_per_second": 150203.86895059218, "data_wait_s": 1.1237003491260111e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 760300, "loss": 0.5664460062980652, "lr": 5e-05, "grad_norm": 0.26154402261595466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111285757983, "tokens_per_second": 152545.6794216428, "data_wait_s": 1.7851001757662743e-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 sha256:151015b2529893d831f50360e31c8aa5e1cd44c21f782c63e5f23a176715a3ca
-size 857865
+oid sha256:cb80f4ec6635a341c438a2effe927f7e6a9089e00e32b9549d9cc48dad596c17
+size 859983